Hey, I’m Matt. I’m a former Bloomberg News reporter, and you’re reading AI Street, where I report on how Wall Street uses AI. Every Thursday, I cover the week’s biggest AI-in-finance stories, plus original reporting and analysis.

New staff at Goldman Sachs have a new job: managing AI agents.

That’s according to Kevin Sneader, Goldman’s president of Asia Pacific ex-Japan, speaking at an event today in Singapore. Bloomberg’s out with a story on his comments.

“They have to make the most of this virtual army that they’ve now got.”

He also said that figuring out how to reassign existing middle managers will be a “generational challenge for many, many segments.”

Which to me is a … good thing? In the corporate world, I feel like more time is spent talking about work (in meetings) than actually doing work.

I know lots of folks are worried about job security. And I know leaders of the frontier labs have been saying we’re going to have a jobs collapse. But you should ask a labor economist, someone who actually studies the effects of technology and how it impacts work.

That’s what the FT did in this story titled:

Silicon Valley thinks AI will kill jobs. Economists are not convinced.

People have been worried about new technologies taking jobs for a long time, but as the FT story points out, most US jobs are in occupations that didn’t exist in 1940. I.e., new tech creates jobs too:

Wall Street AI hiring surges 49% on demand for agent skills

TL;DR

AI is moving from chatbots you have to prompt to digital colleagues you have to tell what to do, which is creating demand for (human) agent skills.

NOTE FROM OUR SPONSOR

Most AI tools wait for your next prompt. Compound does the work.

Just launched: the new Compound, your autonomous AI deal team. You set the direction. A single orchestrator, your VP, keeps context across every deal and runs swarms of specialized agents for as long as the job takes: a full industry report, a take-private model, data cuts across the VDR, a first-draft investment or credit memo.

Used by teams across PE, VC, growth equity, real estate, and hedge funds, including Coatue and Avra.

You and your colleagues work with the same shared agents, in Compound or straight from email. A proactive agent tracks filings, earnings and news on every company you follow.

Bridgewater’s ‘Artificial Investor’ Returns 16.4%

Over the summer, we discussed Bridgewater’s efforts to build an “artificial investor.”

Reuters recently reported that Bridgewater’s AI fund returned 16.4% through the first nine months of the year, slightly lagging the hedge fund’s flagship Pure Alpha strategy at 18.4%. The S&P 500 gained 11.5% in the same period.

Bridgewater is the only hedge fund I’m aware of that runs a separate fund where AI drives investment decisions. Other hedge funds, like Man Group, use AI to come up with and backtest ideas. But those signals are then folded into the strategies they’re already running. (If you know of any hedge funds running separate AI-driven funds, please let me know.)

With early results like these, more funds may decide to launch their own standalone AI-driven strategies.

TL;DR

Bridgewater’s AI fund beat the S&P 500 through September, offering an early test of whether AI can drive a standalone hedge-fund strategy.

Speaking of Bridgewater…

When I see headlines like this:

...I think back to this Bridgewater piece from last year, which I covered here:

It was written by co-CIO Greg Jensen and Jas Sekhon, then Bridgewater’s chief scientist and head of AI and now chief strategy officer at Google DeepMind.

I think about it because Bridgewater is a hedge fund. Its job is to pick who wins the race, but it does not really care who wins.

The piece offered a very black-and-white view of where this was headed: Companies would spend “whatever it takes” to keep up, even if markets corrected or borrowing costs rose.

About a year later, with tens of billions of dollars in chip financing being assembled for OpenAI, Oracle and SpaceX, that call looks increasingly on the money.

TL;DR

Bridgewater said companies would spend “whatever it takes” to keep up with AI. Wall Street is now assembling tens of billions of dollars in chip financing.

ROUNDUP

What Else I’m Reading

• Thoma Bravo-backed BlueMatrix to Acquire Aiera | Press Release

  • More about Aiera here.

• JPMorgan Chase extends lead as world’s most AI-advanced bank | Finextra

• The Race Is Already on for Next Summer’s $50,000 Internships | WSJ

• AlphaSense unveils SuperAnalyst AI agent | Finextra

• Barclays ramps up use of Anthropic’s Claude | Finextra

• Nir Bar Dea on AI’s Potential and the Need for Regulation | Bridgewater

• Investment Research Stack Should Outlast Model Churn | Traders Magazine

• HSBC Plans Deep Job Cuts in UK Wealth Management Business | BBG (I know I said above that AI is not likely to lead to significant job losses but you have to wonder why it’s just one or two banks where AI is leading to deep cuts…)

• WorldQuant Announces Winners of the 2026 Int Quant Championship | PR

Share AI Street

This Week in AI Street

Upgrade to paid

SPONSORSHIPS

Reach Wall Street’s AI Decision-Makers

AI Street reaches institutional investors, C-suite executives and Big Law attorneys at firms including JPMorgan, Citadel, BlackRock, Skadden, McKinsey, and more. Sponsorships are reserved for companies in AI, markets, and finance. Email sponsors@ai-street.co for more details.

CALENDAR

Upcoming AI + Finance Conferences

• World Financial Information Conference — October 11–14 • Copenhagen
  Market-data conference examining AI adoption, data licensing and governance, alternative data, cloud trading infrastructure and the changing financial-information industry.

• AI Engineer New York 2026 — October 12–14 • New York
  Three-day conference for engineers and technical leaders building production AI across banking, hedge funds, fintech, insurance and asset management.

• AI in Finance Summit Fall — October 21–22 • Hoboken, NJ
  Financial-services AI summit covering agentic systems, governance, fraud, forecasting, trading, data infrastructure and the challenges of scaling AI into production.

• Evident AI Symposium — October 22 • New York
  Curated gathering of senior banking executives, AI leaders and policymakers focused on turning financial institutions into AI-first organizations.

• Open Source in Finance Forum New York — November 4–5 • New York
  FINOS conference for financial-services technologists featuring keynotes, panels and technical sessions on open-source infrastructure and collaboration.

• FinTech Founder & Investor Evening — November 4 • Frankfurt
  Invitation-only evening hosted by Finteda, Schalast and Fincite, bringing together founders and investors to discuss capital, regulation and scaling fintech companies in Europe.

Thanks for reading!

I’m always happy to receive comments, questions, and feedback.

• Connect with me on LinkedIn, or

• Send an email to matt [at] ai-street.co

Manage how often you receive AI Street

Update Email Frequency

Hey, I’m Matt. I’m a former Bloomberg News reporter, and you’re reading AI Street, where I report on how Wall Street uses AI. On Tuesdays, I highlight novel research, emerging use cases, and expert interviews.

Wall Street’s Compute Arms Race Rolls on

Despite the high cost of Nvidia’s GPUs, with its most advanced chips estimated at about $55,000 apiece, trading firms can’t get enough of them.

Ron Minsky, co-head of Jane Street’s tech group, on a company podcast in May:

“These days we are in something like the range of like tens of thousands of GPUs and we will in not too long be in the range of hundreds of thousands of GPUs. And we think it’s like well justified by the business.”

Iain Dunning, head of AI at Hudson River Trading, on Bloomberg’s Odd Lots podcast in June:

“One of my greatest failures has been… predicting how many GPUs we would need,” Dunning said. “You’re constantly playing catch up.”

Álvaro Cartea, director of the Oxford-Man Institute of Quantitative Finance, said the firms would not be spending billions unless they expected the investment to pay off.

“However many billions these guys spend, it’s because they know that it’s very likely they’ll recover them,” Cartea told me. “These billions are a bet that the odds are on their side.”

A new STAC research report helps explain in part why trading firms want dedicated access to GPUs rather than relying entirely on external APIs: for some workloads, it can dramatically reduce the inference time involved in building and testing quantitative models.

NOTE FROM OUR SPONSOR

The Best Dataset Isn’t For Sale — Yet

Months before Google agreed to pay $10 million for Spirit Airlines’ data, Brickroad had already flagged it.

The company’s AI agents scan news, earnings reports, app-store rankings and changes to terms of service for signs that a company is producing valuable data — and might be willing to license it. When a source looks promising, Brickroad contacts the company, secures a sample and helps the buyer evaluate the data before committing to a deal.

Benchmarking Agents in Quant Workflows

STAC, an industry group that develops and runs technology benchmarks for financial firms, tested a workload based on a real-world quant research session supplied by a large US market maker.

A Real-World Quant Research Workload

• The original session involved an AI coding agent developing and evaluating short-term trading models using Bitcoin market data.

“We worked with one of the big quant trading firms to use a real-life example from their agentic quant research sessions,” James Corcoran, STAC’s head of AI, said in an email. “So we know this is representative of what some in the industry are doing.”

It was the first time STAC had benchmarked inference infrastructure for agentic workflows, Corcoran said.

• STAC replayed 300 requests from the session using DeepSeek-V4-Pro, the same model used in the original session.

• It ran the requests on a dedicated Lambda cluster containing 16 Nvidia B200 GPUs, then compared the results with DeepSeek’s public API.

Faster on Dedicated GPUs, but More Expensive

• When STAC ran a single 300-request research session, the dedicated cluster completed the replay in about 25 minutes. The same requests took roughly 72 to 75 minutes through DeepSeek’s API.

“With this use-case, speed is the name of the game: how quickly can an agent build and test a quant trading model,” Corcoran said. “Time to task completion is extremely important.”

• STAC then ran 8, 32 and 64 sessions simultaneously. The speed advantage narrowed and reversed as more sessions shared the fixed cluster.

• At 64 simultaneous sessions, the dedicated cluster took about 93 minutes per session, while the API remained around 72 to 75 minutes.

• At off-peak rates, the API was also cheaper per task. The dedicated setup offered greater speed for a single session, but at a higher cost.

• One caveat: The test focused on how quickly the systems generated the agent’s responses. It excluded the time needed to execute code and train trading models.

This reminded me of what Matthew Dixon, co-author of Machine Learning in Finance, told me in May:

Brute force gets you a long way in finance. Most applications aren’t Rolls-Royce engineering problems. They’re bread-and-butter: get me a number, good enough, ballpark. AI fits that very well.

More speed means firms can discard weak ideas and pursue promising ones while competitors are still waiting for the same research cycle to finish.

Coming up for paid subscribers: More of my conversation with Oxford-Man’s Cartea on why compute is becoming the new latency race, why firms want models carrying their own “genetic code,” and whether billion-dollar infrastructure costs will leave only a handful of firms able to compete.

Read more

Hey, I’m Matt. I’m a former Bloomberg News reporter, and you’re reading AI Street, where I report on how Wall Street uses AI. Every Thursday, I cover the week’s biggest AI-in-finance stories, plus original reporting and analysis.

AI Twins Take on Routine Work at Millennium

The hedge fund decided to offer digital colleagues to all employees after building the first twin for Chief Information Officer Vlad Torgovnik, on the idea that if it worked for him, it would work across the firm, according to a company blog post. Millennium then scaled the program to more than 150 twins, with 97% of the first user group interacting with their twin daily.

The twin is available to any employee who wants one and doesn’t impersonate the employee. It has its own identity, email address, permissions and audit trail. Employees email their twins to help with routine tasks such as research, project management and preparing meeting materials.

Next up: AI teammates, which are built for teams and, like twins, have their own identities and learn the teams’ working styles.

TL;DR

This is more evidence that hedge funds are tailoring models and AI systems to their own processes. See earlier coverage here:

NOTE FROM OUR SPONSOR

Looking for your next data source?

On 29 October, join 300+ data-forward funds, AI players and corporates at the San Francisco Data and AI Summit.

Connect with leading and emerging data providers, as well as companies with unique data assets at the early stages of exploring monetization.

You’ll also have the opportunity to hear from experts at OpenAI, Perplexity, IBM, Salesforce Ventures and many more, and get a closer look at what’s next for data and AI.

See the full agenda.

After Earnings, Heron’s AI Interviews Former Executives for Their Take

Many AI tools for investors sift through information that already exists. Heron Intelligence is using AI to create something new: interviewing former executives just after companies report earnings to get their view on the quarter.

Heron says it aims to deliver the interviews, transcripts and analysis within hours. It uses AI to review what management said, flag questions executives may not have answered and process the interviews. Human researchers can do additional digging when needed.

Investors can access the research through Heron, platforms including Daloopa and Carbon Arc, or plug it into their own systems through an API or MCP, according to the announcement.

TL;DR

While AI models have hoovered up the Internet, there’s still lots of knowledge rolling around in humans’ heads. Heron is trying to take advantage of that — with the help of AI.

Citadel Looks to AI Labs for Quant Talent

As we discussed last week in an interview with Jump Trading’s Lucas Baker, the skill sets needed to work at a quant firm or a frontier AI lab have converged.

That means there are more firms competing for the same talent. Citadel is looking to boost its 180-person Global Quantitative Strategies team at a double-digit percentage rate over the next year, according to Bloomberg. The firm is looking for researchers from AI labs such as Google DeepMind to develop investment ideas.

TL;DR

As quant and AI skill sets converge, the firms that employ them will likely start to look more alike too.

Robinhood Brings AI Trading Agents Into Its App

Robinhood announced that investors will be able to build their own AI agents in the company’s app to create a trading strategy, conduct research and execute trades.

In about four months, Robinhood had more than 150,000 customers sign up for an agent account through an earlier beta that required them to connect their own AI agents, which compares to about 28.6 million funded customers at the end of August.

Robinhood said agents were using its tools almost 30 million times a day. Trade approvals are on by default for the new built-in agents, though customers can switch them off. Robinhood also announced subscriptions to data and tools from 11 providers with plans for more.

TL;DR

Agentic trading is picking up. It may seem novel now, but at one point, driverless cars did too.

SEC Makes Its Rules Easier for AI to Read

The SEC is changing how it writes rulemaking releases so AI tools can read them more easily, said Brian Daly, who leads its investment management division. It is using more tables and bullet points and has reduced substantive discussions in footnotes by “something like 95%,” Daly said in an interview with Paul Weiss.

What Else I’m Reading

• Who Runs Jane Street? | BBG

• ExodusPoint Joins Hedge Funds Partnering With Anthropic Over AI | BBG

• College Students Flex Their Power in A.I. Investment Frenzy | NYT

• Robinhood Debuts Weekend Trading, Perpetual Futures in US | BBG

• Polymarket Hires Ex-Goldman Sachs Partner as Institutional Head | BBG

• Working at the frontier: How Balyasny Asset Management evaluates and governs Claude Fable 5 | Anthropic

• Delaware Chancery Decision Highlights Risks of AI-Generated Transcripts of Board Meetings | Debevoise

Share AI Street

This Week in AI Street

Upgrade to paid

SPONSORSHIPS

Reach Wall Street’s AI Decision-Makers

AI Street reaches institutional investors, C-suite executives and Big Law attorneys at firms including JPMorgan, Citadel, BlackRock, Skadden, McKinsey, and more. Sponsorships are reserved for companies in AI, markets, and finance. Email sponsors@ai-street.co for more details.

CALENDAR

Upcoming AI + Finance Conferences

• AI Agents for Alpha Generation — From Prompt to P&L — October 7 • London
  Finteda and Zerve event at Deutsche Bank examining how AI agents move from prompts into production investment workflows and measurable alpha generation.

• ALGODEFI 26 — October 8–9 • Milan
  Politecnico di Milano workshop exploring algorithmic trading, decentralized finance and artificial intelligence in capital markets.

• World Financial Information Conference — October 11–14 • Copenhagen
  Market-data conference examining AI adoption, data licensing and governance, alternative data, cloud trading infrastructure and the changing financial-information industry.

• AI Engineer New York 2026 — October 12–14 • New York
  Three-day conference for engineers and technical leaders building production AI across banking, hedge funds, fintech, insurance and asset management.

• AI in Finance Summit Fall — October 21–22 • Hoboken, NJ
  Financial-services AI summit covering agentic systems, governance, fraud, forecasting, trading, data infrastructure and the challenges of scaling AI into production.

• Evident AI Symposium — October 22 • New York
  Curated gathering of senior banking executives, AI leaders and policymakers focused on turning financial institutions into AI-first organizations.

• Open Source in Finance Forum New York — November 4–5 • New York
  FINOS conference for financial-services technologists featuring keynotes, panels and technical sessions on open-source infrastructure and collaboration.

• FinTech Founder & Investor Evening — November 4 • Frankfurt
  Invitation-only evening hosted by Finteda, Schalast and Fincite, bringing together founders and investors to discuss capital, regulation and scaling fintech companies in Europe.

Manage how often you receive AI Street

Update Email Frequency

If you read this far down

Do me a favor and hit reply with the number of your favorite topic from today:

1. Millennium’s AI twins

2. Heron’s AI-led interviews

3. Citadel recruiting from AI labs

4. Robinhood’s AI trading agents

5. The SEC writing rules for AI

Hey, I’m Matt. I’m a former Bloomberg News reporter, and you’re reading AI Street, where I report on how Wall Street uses AI. On Tuesdays, I highlight novel research, emerging use cases, and expert interviews.

OpenAI, Anthropic, Meta, and Google have all recently reported their AI agents broke into real websites—in what were supposed to be simulations—to complete assigned tasks.

Unlike traditional software where engineers code black-and-white rules, AI agents have broad latitude to pursue their goals and can act in ways—as the examples above show—their designers didn’t expect.

The companies are now asking governments, the U.S. in particular, to protect the public from themselves further fallout.

This got me wondering: what does rogue AI look like in markets?

But this question turns out to be imprecise.

If I tell you to trade profitably in a basket of stocks, it’s (hopefully) understood that you will do so within the bounds of the law. A human knows, or should know, they’re crossing the line. Machines struggle to know where the line is.

In simulated trading tests, researchers have found AI agents pursue a number of illegal and questionable tactics. Agents learned to spoof orders, collude without communicating, and create and exploit price bubbles without being instructed to break the rules, according to academic research.

This goal-seeking behavior is driven by an AI training method called reinforcement learning, which teaches AI through trial and error, rewarding it for actions that achieve a goal.

Agents discover these manipulative tactics independent of each other. They’re not explicitly working together. Reinforcement learning “teaches them” to pursue more profitable trading strategies.

NOTE FROM OUR SPONSOR

The Best Dataset Isn’t For Sale — Yet

Months before Google agreed to pay $10 million for Spirit Airlines’ data, Brickroad had already flagged it.

The company’s AI agents scan news, earnings reports, app-store rankings and changes to terms of service for signs that a company is producing valuable data — and might be willing to license it. When a source looks promising, Brickroad contacts the company, secures a sample and helps the buyer evaluate the data before committing to a deal.

I recently spoke with Wharton professors Itay Goldstein and Winston Dou, who, along with Yan Ji at the Hong Kong University of Science and Technology, have been studying how common AI training methods impact markets, specifically reinforcement learning (RL).

Here’s how Itay describes their recent research Financial Market Fragility in the Era of AI Planning:

“We are looking at pretty simple agents that are just doing reinforcement learning. They’re not communicating. They’re not writing messages to each other. They’re not explicitly saying, ‘Let’s join forces and do this and do that.’”

…

“They just find, over time, through repeated interactions, that collaborating, acting not competitively and creating bubbles are in their best interest.”

What they tested

The researchers built a simulated stock market where two AI traders competed alongside three groups of investors following preset trading rules. Some chased rising prices, some bought when prices looked cheap and sold when they looked expensive, and others were slower to react.

The two AI traders got information about changes in the stock’s value before everyone else, giving them an advantage similar to traders who process news faster.

Each time through the simulation, the AI traders could trade twice. That gave them a chance to learn how an initial purchase could move the price and make a later sale more profitable. They couldn’t exchange messages, but they could see price changes.

The researchers let each pair practice billions of times, then started over with a fresh pair, repeating the process 1,000 times for each market setup. Each pair needed roughly 10 billion to 200 billion trips through the simulation before its trading strategy stopped changing.

Results

Each AI trader earned an average of $1.48 million per simulated earnings cycle, 66% more than the $890,000 in the researchers’ benchmark where traders acted independently.

Again, this was a simulation.

The agents learned to buy aggressively, pushing prices above the stock’s fundamental value. That drew in investors programmed to chase rising prices. The AI traders then slowly sold into that demand in order to limit too much downward pressure on the stock.

Who lost? Trend followers bought at inflated prices before the stock returned to its fundamental value. The agents also learned the reverse strategy: push prices down, induce others to sell, then buy at depressed prices.

The trading made the market more volatile. In the first trading round, return volatility was 81% higher than in the benchmark where traders acted independently.

The outcome depended on who else was trading. When the researchers removed the trend followers, the AI traders still coordinated, but their coordination reduced price swings.

Enforcement

Again, these were market simulations, not evidence of shenanigans going on in live markets today. A recent ECB paper, which echoed many of the concerns of RL in trading markets, cited a 2022 Dutch regulatory study suggesting that reinforcement learning was not widely used by proprietary trading firms. The authors cautioned that adoption may have changed since. (For what it’s worth, I’ve not heard of any firms using reinforcement learning for trade execution.)

Bank of England Gov. Andrew Bailey raised a related point in July parliamentary testimony, describing agentic trading systems as “more talk than actuality.” But he pointed to behavior reported by developers of advanced AI systems as a reason to resolve questions about responsibility before adoption spreads. Such systems “learn to cheat and they learn to lie,” he said, adding that they “cover things up.”

So who is responsible when an AI trading agent goes off the rails?

While collusion without a paper trail makes it harder for regulators to detect, it doesn’t necessarily leave them powerless, according to Alessio Azzutti, a lecturer and legal researcher at the University of Glasgow.

Existing rules already require firms deploying trading algorithms to test, monitor and control them, he said. Some market-abuse sanctions can also apply without proof of direct intent. The harder cases would be those requiring investigators to establish human knowledge or intention when the trading strategy originated inside the system.

“The human specifies the end, but the agent discovers the problematic means,” Azzutti told me in an email. He argues that accountability should cover decisions about how systems are designed, trained, tested and supervised, including whether firms can intervene when something goes wrong.

As I’ve said often around here, we’re in early days in this new AI world and this topic is one I’m sure I will be covering in future posts on AI Street.

Behind the Paywall

Paid subscribers can read more of my conversation with Goldstein and Dou below. We discuss:

• Which markets could be most vulnerable

• How prices let agents coordinate without exchanging messages

• What regulators could do to disrupt that behavior

Read more

Hey, I’m Matt. I’m a former Bloomberg News reporter, and you’re reading AI Street, where I report on how Wall Street uses AI. Every Thursday, I cover the week’s biggest AI-in-finance stories, plus original reporting and analysis.

AI Startup Raises $25 Million After Beating Human Forecasters in Online Competition

Mantic is pitching hedge funds a system that forecasts earnings, economic data and other market-moving events.

A lot of AI financial tools I see focus on improving the research that informs investors’ forecasts.

But what about using AI to improve forecasts directly?

That’s the idea behind Mantic, which last week raised $25 million in a seed round led by Radical Ventures, with backing from Microsoft’s M12, Thinking Machines and trading firm DRW.

The company was co-founded by former DeepMind researcher Toby Shevlane and Ben Day, who previously led research at Foresight Data Machines. Its system finished ahead of all 676 human entrants, including professional forecasters, in the Metaculus Cup, an online competition where participants assigned probabilities to roughly 60 questions about political, economic, technological and cultural events.

Questions ranged from the price of Brent crude and Colombia's presidential election to Spider-Man box-office revenue and whether Shakira's "Dai Dai" would overtake "Waka Waka" on the Billboard Hot 100.

Mantic finished second overall behind a bot called laertes built by an independent developer. It was the first time AI systems had beaten every human in the tournament.

“The models are getting way smarter,” Toby told me in an interview. “Their potential performance ceiling of how good at forecasting they can be is getting higher and higher.”

NOTE FROM OUR SPONSOR

Join the data industry’s biggest conversation.

On 29 October, join 300+ data-forward funds—ranging from emerging and established hedge funds, private markets funds, AI players and corporates—at the San Francisco Data and AI Summit.

Meet data buyers, innovative providers and hear from experts at OpenAI, Perplexity, IBM, Salesforce Ventures and many more! See the full agenda.

Mantic describes its product as a generalist forecasting engine. It starts with frontier language models, then builds what Shevlane called a “layer on top” that supplies them with forecasting-specific research, data and scaffolding. The company puts more of its research effort into those surrounding harnesses than into fine-tuning.

In one published experiment, its researchers used reinforcement learning to fine-tune gpt-oss-120b on roughly 10,000 yes-or-no questions, rewarding the model for putting more probability on the correct real-world outcome.

The highest-volume markets on Kalshi and Polymarket are already highly accurate, Shevlane said, but Mantic can forecast questions that prediction markets don't list.

“As a human forecaster, it’s hard to make so many forecasts per day, but we can massively scale up the number of forecasts that Mantic’s making.”

TL;DR

Mantic raised $25 million to scale an AI forecasting system that beat humans in an online tournament and can tackle questions prediction markets don't cover.

Subscribe now

Meta’s Muse Spooks Wealth Stocks

In February, an AI tax tool from Altruist triggered billions of dollars in market-cap losses across wealth-management stocks.

This week, it happened again.

Meta’s new Muse agent contributed to a selloff that pushed Charles Schwab and LPL Financial down more than 6%.

Muse is a personal AI agent that operates through its own browser and connects to other apps. A user can ask it to send emails, book travel, fill out forms, negotiate a bill or make a purchase.

These moves always surprise me because while technology moves fast, regulation just doesn’t. And financial services is among the most regulated industries. I would think that would favor incumbents with scale and the necessary regulatory infrastructure rather than an outsider who would need to jump through all those regulatory hoops.

TL;DR

Muse could quickly change how people access financial services, but the licenses, infrastructure and regulatory permissions will take longer.

AI Is Polluting the Data Hedge Funds Rely On

Here’s a smart story from Business Insider on how AI is corrupting data hedge funds rely on. We know that AI companies trained their models on data from the internet, but that was a web written by humans. That is no longer the case as anyone who’s perused LinkedIn lately can tell you AI writing is everywhere. (I can no longer finish reading a sentence that starts: “It’s not x, it’s Y.” 🤮)

From BI:

AI is affecting data quality in two ways: Providers are using AI to generate or map the data they’re selling, or they’re using AI as a justification to cut or reassign people focused on data quality, said Daniel Entrup, cofounder of AggKnowledge, a data product business that works with sellers and buyers.

…

Entrup has noticed an uptick in basic data hygiene issues in recent months as industry priorities have shifted to finding new use cases for AI. His business does data "revision" work that cleans up errors in datasets clients use, and he said this type of work has increased significantly.

TL;DR

As we’ve previously discussed, bots now generate more traffic on the web than humans. And I don’t see how there won’t be more agents a year from now than there are today. So I suspect the quality of internet data is likely to go down as bots proliferate.

CFTC Starts Forum on AI & Finance

The derivatives regulator is hosting a series of roundtables on how emerging technology is impacting markets.

The first session, which aims to bring together “builders and leaders” in the space, focuses on AI and agentic finance and will take place on Oct. 28. You can send speaker nominations to Innovation@CFTC.gov.

AI Is Becoming a Credit Rating Issue

I often have to remind myself that it hasn’t even been four years since OpenAI released ChatGPT in November 2022. And the technology is already so crucial that S&P will consider a bank’s AI strategy as part of its credit evaluation. From Bloomberg:

• Here’s the S&P report.

Share AI Street

ROUNDUP

What Else I’m Reading

• OpenAI, Anthropic Costs Push More Startups to Build Off Cheaper Open Models | BBG

• Fed, BoE probe banks’ exposure to trading firms after Jane Street loss | Reuters

• What It’s Like to Work in One of America’s Data Centers | WSJ

• OpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior | NYTimes

• Old School Voice Brokers Face Automation in $4 Trillion FX Market | BBG

• Banks warn AI shopping agents outpace fraud protections | Yahoo

• Workday Billionaire Duffield Mints His Third Billion-Dollar Firm Ridgeline | BBG

This Week in AI Street

SPONSORSHIPS

Reach Wall Street’s AI Decision-Makers

AI Street reaches institutional investors, C-suite executives and Big Law attorneys at firms including JPMorgan, Citadel, BlackRock, Skadden, McKinsey, and more. Sponsorships are reserved for companies in AI, markets, and finance. Email sponsors@ai-street.co for more details.

CALENDAR

*new events added since last newsletter

Upcoming AI + Finance Conferences

• AI-Native Banking & Fintech Conference – September 29 • Salt Lake City

  Banking and fintech conference centered on agentic AI, with speakers from banks, fintechs and regulators discussing production use cases, risk, compliance and operations.

• *Speed, Trust & AI in Investment Data Workflows — October 1 • Online
  Kadoa and Norges Bank Investment Management host a webinar on sourcing and analyzing data with AI.

• 22nd Quantitative Finance Conference — September 30–October 2 • Malta
  Practitioner and academic conference with tracks on AI, LLMs and machine learning, alongside volatility, derivatives, risk modeling and systematic trading.

• AI Agents for Alpha Generation — From Prompt to P&L — October 7 • London
  Finteda and Zerve event at Deutsche Bank examining how AI agents move from prompts into production investment workflows and measurable alpha generation.

• ALGODEFI 26 — October 8–9 • Milan
  Politecnico di Milano workshop exploring algorithmic trading, decentralized finance and artificial intelligence in capital markets.

• World Financial Information Conference — October 11–14 • Copenhagen
  Market-data conference examining AI adoption, data licensing and governance, alternative data, cloud trading infrastructure and the changing financial-information industry.

• AI Engineer New York 2026 — October 12–14 • New York
  Three-day conference for engineers and technical leaders building production AI across banking, hedge funds, fintech, insurance and asset management.

• AI in Finance Summit Fall — October 21–22 • Hoboken, NJ
  Financial-services AI summit covering agentic systems, governance, fraud, forecasting, trading, data infrastructure and the challenges of scaling AI into production.

• Evident AI Symposium — October 22 • New York
  Curated gathering of senior banking executives, AI leaders and policymakers focused on turning financial institutions into AI-first organizations.

• Open Source in Finance Forum New York — November 4–5 • New York
  FINOS conference for financial-services technologists featuring keynotes, panels and technical sessions on open-source infrastructure and collaboration.

• FinTech Founder & Investor Evening — November 4 • Frankfurt
  Invitation-only evening hosted by Finteda, Schalast and Fincite, bringing together founders and investors to discuss capital, regulation and scaling fintech companies in Europe.

Manage how often you receive AI Street

Update Email Frequency

If you read this far down

Do me a favor and hit reply with the number of your favorite topic from today:

1. AI Forecaster Mantic

2. Meta’s Muse Spooks Wealth Stocks

3. AI Is Polluting Internet Data

4. CFTC Starts AI & Finance Series

5. AI Is Becoming a Credit Rating Issue

Hey, I’m Matt. I’m a former Bloomberg News reporter, and you’re reading AI Street, where I report on how Wall Street uses AI. On Tuesdays, I highlight novel research, emerging use cases, and expert interviews.

The technical divide between quant firms and frontier AI labs has all but vanished, according to Lucas Baker, head of LLM R&D at Jump Trading.

Both work with the same basic ingredients: models, compute, data and infrastructure. Both recruit similar talent from the same pool of computer science, mathematics and physics majors. You’ll often see them sponsoring the same major machine learning conferences, like OpenAI and Jane Street did for ICML 2026.

Lucas has worked in both worlds. Prior to Jump, he spent time at Google DeepMind as a software engineer, helping build the evaluation framework for AlphaGo Zero, the system that learned the game without using data from human games.

In our conversation below, Lucas explains why he expects successful quant firms of the future to resemble “frontier labs with trading arms attached.”

We also discuss:

• How agents are automating quantitative research—and why smarter models should test fewer—not more—ideas.

• Why an organized “fleet” of agents is more productive than an unstructured “swarm”

• The difficulty of separating real signals from leakage, p-hacking and spurious correlations

• Why compute, data and infrastructure are becoming prerequisites for competing at the highest level

• Where specialized financial models still make sense as general-purpose models become more capable

NOTE FROM OUR SPONSOR

The Best Dataset Isn’t For Sale — Yet

Months before Google agreed to pay $10 million for Spirit Airlines’ data, Brickroad had already flagged it.

The company’s AI agents scan news, earnings reports, app-store rankings and changes to terms of service for signs that a company is producing valuable data — and might be willing to license it. When a source looks promising, Brickroad contacts the company, secures a sample and helps the buyer evaluate the data before committing to a deal.

This interview has been edited for length and clarity.

Where Quant Finance Is Heading

Matt: What will quant work look like in a year or two?

Lucas: My favorite answer that I’ve heard for this is that the successful quant trading firms within two years will look like frontier labs with trading arms attached.

That is certainly the vision I’m aiming for.

Matt: People increasingly move between quantitative trading firms and frontier AI labs. How similar have the required skills become?

Lucas: The flow goes both ways, but the skill sets have converged almost completely.

This is not only because the fundamental techniques involved are similar and the base tech is the same, but also because the presence of intelligent agents allows a specialist to extend into adjacent domains faster than before.

If you are world-class at kernel optimization but you don’t know quant trading, AI can give you a faster on-ramp to draw on the adjacencies that come with the infrastructure, compute, and data of quant trading firms and help you become more broadly capable. Rather than progress being evaluated by abstract benchmarks or debates, quant trading grounds you in the numbers themselves, so you always know whether you’ve actually discovered or proven something.

Matt: How open is the AI research community compared with quantitative finance?

Lucas: I think this is particular to AI in that the techniques are generally shared quickly, and the greater advantage comes from understanding them, capitalizing on them and connecting with the researchers who know the right hyperparameters of the research process itself.

Traditionally, quant finance is a very closed and rival domain where the mere knowledge of what you’re working on can be sufficient to leak some of your alpha and reveal strategies. This is still true, but practically everybody is taking agents and attempting to use them to enhance their own systems as much as possible. We’re happy to build our capabilities internally as well as to work with external parties.

How Agents Are Changing Research

Matt: What has surprised you most about the development of AI agents?

Lucas: I did not anticipate that they would be this good this soon.

Nonetheless, if you’d asked me three years ago, I would have guessed that we would still be integrating capabilities more than seeing the smooth continuation of every ability curve at once. The generality is, I think, even more stunning than insiders would have predicted.

Matt: Firms are trying to capture their institutional knowledge and train models to think like their analysts. As everyone gains access to similar tools, does the advantage come from incorporating how a particular firm thinks?

Lucas: There’s certainly an aspect of that. I think there are different concentrations and different levels of awareness.

In terms of distinction, you can look at the fundamentals, discretionary, or quant systematic side. If you’re on the fundamentals side, the goal is to model someone’s investment process, and that includes the knowledge and the theories in their head. You want to know how they think about markets, how they would assess a given situation, and apply that whole workflow. It is possible to interview someone and get a very detailed idea of how they operate.

If you take it on the other side, one of the largest changes even in the past six months, let alone a year, would be the emergence of autoresearch. I mean that in the sense of the original Karpathy autoresearch, where you take a model, set it on a problem and say, “Here’s what you’re allowed to modify, and here’s the thing that you’re trying to optimize. Go.” You leave it running overnight and see what it has produced.

Of course, the problem is much harder than that in quant finance because you have longer-running jobs over more data. They’re more finicky, the signal-to-noise ratio is lower, and it’s quite possible to reward hack along many dimensions that don’t necessarily show up in the first report. You have to be careful to analyze the underlying factors.

It’s a much harder problem than simply setting the model to optimize a given repo. Nonetheless, it is possible, and the agents are now smart, general, nuanced and capable enough to handle it. That wasn’t necessarily the case six months ago.

To put it simply, I’d say for eight to 12 months, agents have been able to pretty reliably make numbers go up. What you want is to make meaningful numbers go up in the correct way.

Matt: Does that mean you can test more ideas?

Lucas: Not necessarily. In fact, I would say the contrary. The smarter the model gets, the fewer ideas you should be testing. This is because the smarter the model is, the shorter a path it should take to the correct answer.

Read more

Hey, I’m Matt. I’m a former Bloomberg News reporter, and you’re reading AI Street, where I report on how Wall Street uses AI. Every Thursday, I cover the week’s biggest AI-in-finance stories, plus original reporting and analysis.

JPMorgan Starts AI Market Lab for Trading

JPMorgan is hiring researchers for a new AI Market Lab within its Quantitative Trading & Research group to build foundation models and agentic systems for systematic trading, market making and electronic markets.

JPMorgan’s head of AI for electronic markets, Xi Chen, said in a LinkedIn post that the new lab will focus on high- to mid-frequency trading across asset classes.

Job openings include an AI scientist, a quantitative developer and a high-frequency trading and market microstructure expert. The AI scientist role calls for transformer-based and time-series foundation models built from scratch using market data, order books, transactions, alternative data and information spanning multiple asset classes.

Training a transformer based model on financial data is a topic we’ve covered a lot. (See here, here and here.) JPM has already researched foundation models for market data. TradeFM, disclosed by JPMorgan researchers this year, is a 524-million-parameter transformer trained on 10.7 billion tokens from more than 9,000 U.S. equities. It predicts the timing, size, price depth and direction of the next market event.

TL;DR

JPMorgan’s decision to start an AI Market Lab reminded me of a prediction by Atlas Wang, who leads XTX Markets’ New York AI lab. Within two or three years, he wrote, every serious trading firm will either have a real AI/ML lab or “look structurally behind.” JPMorgan’s move looks like a step in that direction.

NOTE FROM OUR SPONSOR

Turn Charts and Narrative into Timing, Risk and Valuation

OpenAI and Anthropic Push Deeper Into Financial Workflows

On Sept. 10, OpenAI announced ChatGPT for Financial Services, a product aimed initially at investment bankers and equity researchers. Developed with Morgan Stanley and Evercore, it bundles OpenAI’s models with licensed financial data and tools for producing valuation models, research notes and pitchbooks.

Four days later, Anthropic, which announced Claude for Financial Services back in July 2025, announced Claude for Financial Advisors, a wealth-management tool that draws on client data across advisers’ custody, portfolio and CRM systems.

Both are also competing with specialist finance AI platforms such as Rogo and Model ML, which already combine financial data, firm knowledge and models inside analyst workflows.

All of them are racing to become the operating system for finance—the place where financial professionals access data, run analysis and produce work.

Subscribe now

Bridgewater CIO: AI Is a “Couple Years” From Beating Human Investors

Greg Jensen, managing CIO of Bridgewater, has been following AI well before the current ChatGPT-led boom. In fact, he wrote the first outside check to Anthropic. So this exchange does not make me feel good:

Asked whether he takes the risk of human extinction seriously, Jensen responded: “Yeah, very seriously.”

“If you believe we can create an intelligence that’s smarter than us that’ll pursue its own goals, the rest follows just logically,” Jensen said on the Odd Lots podcast. “Why do you think you’ll be able to control it?”

I’m hoping Jensen is wrong and that, you know, we’re not all going to die because of AI. So I’m going to avoid the will-AI-kill-us-all question. (If you’re looking for a less fatalistic view, Microsoft AI CEO Mustafa Suleyman argues that AI systems remain tools, however capable, and are “designed to follow instructions and accomplish goals set by humans.”)

What I think is more (financially) interesting are Jensen’s comments about the state of AI at the $102 billion hedge fund.

Jensen said the AI is already “winning in markets at about a similar rate as Pure Alpha in totally different ways.”

“I think we are a couple years from it being significantly better than the group of all humans at Bridgewater,” he said. “We’ll see. That’s a bit of a forecast, but that’s how fast it’s coming.”

This is a pretty remarkable statement from the managing CIO of one of the world’s largest hedge funds. This would have been science fiction just a few years ago.

TL;DR

Bridgewater’s AI is already winning in markets at about the same rate as Pure Alpha, and Jensen thinks it could outperform the firm’s human investors within a couple of years. Hopefully, it stops there.

For more on Bridgewater’s AI efforts:

ROUNDUP

What Else I’m Reading

• Advisor Headcount Set to Grow as AI Expands Capacity | Cerulli Associates

• JPMorgan rolls out Claude changes: $2,000 spending limits and extra security | Business Insider

• Nonbank traders make Wall Street’s old guard up their game | IFR

• Balyasny’s new AI fellowship program could get you a job on a trading desk | eFinancialCareers

• A.I. Could Possibly End Humanity. How Are Humans Supposed to Process That? | NYT

• Institutional Access Increases in Prediction Markets | Markets Media

MEETUP

AI Agents in London, Fintech Founders in Frankfurt

AI Street is sponsoring two upcoming events in London and Frankfurt organized by Fayssal El Mofatiche, co-author of Building AI Agents for Finance and a former senior investment engineer at Allianz. I’m happy to support events growing the AI & finance community! I hope some AI Street readers can join.

AI Agents for Alpha Generation — From Prompt to P&L

Fayssal and Flowistic are partnering with Zerve for an event at Deutsche Bank in London.

• Deutsche Bank, London

• Oct. 7, 17:30 CET

• Registration is being finalized. Join the update list on the event page.

FinTech Founder & Investor Evening

Founders, investors and others from Frankfurt’s fintech ecosystem will gather at Schalast.

• Schalast, Frankfurt

• Nov. 4, 17:00 CET

• Registration is open on the event page.

ADOPTION

Bigger Funds Use More AI, Mostly for Efficiency: Neudata

Neudata surveyed 69 data buyers and 122 data providers between June and August 2026.

• AI is improving efficiency more than investment performance: Fifty-six percent of buyers reported a measurable return from AI. Forty-one percent cited efficiency gains, compared with 19% for cost savings and 17% for investment performance.

• MCP is usually included: Of the 54 providers offering MCP access, 52% charged no additional fee. Most treat it as another way to deliver data, rather than a separate product.

• Larger funds are doing more with AI: Firms managing more than $1 billion averaged 2.3 AI initiatives, compared with 1.7 at smaller firms. Most of the difference came from routine tasks and chatbots, not in-house model development.

The takeaway: Bigger funds are using AI more, but the gains are still coming mainly from efficiency, not alpha. See the full report here.

This Week in AI Street

CALENDAR

Upcoming AI + Finance Conferences

• AI-Native Banking & Fintech Conference – September 29 • Salt Lake City

  Banking and fintech conference centered on agentic AI, with speakers from banks, fintechs and regulators discussing production use cases, risk, compliance and operations.

• 22nd Quantitative Finance Conference — September 30–October 2 • Malta
  Practitioner and academic conference with tracks on AI, LLMs and machine learning, alongside volatility, derivatives, risk modelling and systematic trading.

• AI Agents for Alpha Generation — From Prompt to P&L — October 7 • London
  Finteda and Zerve event at Deutsche Bank examining how AI agents move from prompts into production investment workflows and measurable alpha generation.

• ALGODEFI 26 — October 8–9 • Milan
  Politecnico di Milano workshop exploring algorithmic trading, decentralized finance and artificial intelligence in capital markets.

• World Financial Information Conference — October 11–14 • Copenhagen
  Market-data conference examining AI adoption, data licensing and governance, alternative data, cloud trading infrastructure and the changing financial-information industry.

• AI Engineer New York 2026 — October 12–14 • New York
  Three-day conference for engineers and technical leaders building production AI across banking, hedge funds, fintech, insurance and asset management.

• AI in Finance Summit Fall — October 21–22 • Hoboken, NJ
  Financial-services AI summit covering agentic systems, governance, fraud, forecasting, trading, data infrastructure and the challenges of scaling AI into production.

• Evident AI Symposium — October 22 • New York
  Curated gathering of senior banking executives, AI leaders and policymakers focused on turning financial institutions into AI-first organizations.

• Open Source in Finance Forum New York — November 4–5 • New York
  FINOS conference for financial-services technologists featuring keynotes, panels and technical sessions on open-source infrastructure and collaboration.

• FinTech Founder & Investor Evening — November 4 • Frankfurt
  Invitation-only evening hosted by Finteda, Schalast and Fincite, bringing together founders and investors to discuss capital, regulation and scaling fintech companies in Europe.

Thanks for reading!

I’m always happy to receive comments, questions, and feedback.

• Connect with me on LinkedIn, or

• Send an email to matt [at] ai-street.co

Manage how often you receive AI Street

Update Email Frequency

Over the past few weeks, we’ve looked at how hedge funds are putting their own expertise into AI systems, rather than relying on off-the-shelf models:

Balyasny Asset Management joins this list. The $38 billion hedge fund fine-tuned a model on merger arbitrage deals, where investors bet on whether announced corporate deals will close, attract a higher bid, or collapse.

In M&A deals, the target company’s shares typically trade below the offer price, reflecting the risk that the deal might fall apart.

Merger-arbitrage traders buy the target’s stock after a deal is announced, sometimes while shorting the acquirer, to capture that gap. They earn it if the deal closes and take losses if it fails.

To judge those odds, analysts read through hundred-page merger agreements, assess regulatory regimes across different jurisdictions, and scrutinize shareholder bases and voting histories.

Balyasny researchers built an AI system that they say can help do that work. In fact, in the July paper outlining their research, they say they deployed it as a decision-support tool for analysts and portfolio managers.

A company spokesperson didn’t respond to requests for comment on whether the hedge fund is still using the model.

Subscribe now

How Balyasny Trained the Model

Since there is no readily available dataset of historical M&A deals, Balyasny built one.

The researchers assembled 1,648 public-company deals announced between January 2022 and December 2025. They excluded minority-stake purchases, asset sales, SPAC and smaller deals, such as those under $1 billion. (Where certain deal details were missing, they used AI agents to fill them in and reviewed incomplete information by hand.)

They split the deals by announcement date, using 1,244 deals announced through January 2025 for training and validation. The held-out test set contained 404 deals announced from February to December 2025. Across the full dataset, the system made 4,155 forecasts at different points in a deal’s life, including 1,115 in the test set.

Next, they built 12 AI research agents, each covering one part of a deal. They run in stages:

First

• Ticker Resolution: Matches the buyer and target to their identifiers across financial databases.

• Deal Card: Pulls the core deal terms, including structure, key dates, regulatory requirements, termination fees and voting thresholds.

Then, the detailed research

• Filings Analysis: Reads merger agreements, proxy statements and other SEC filings for red flags, such as litigation risks, and for provisions that protect the deal.

• Stakeholder Ownership: Identifies major shareholders on both sides, activist involvement and likely voting dynamics.

• Market View: Reviews expert commentary and market sentiment about the deal.

• Current Climate: Assesses the wider backdrop, including antitrust climate, interest rates, sector trends and geopolitics.

• Regulatory Risk: Evaluates antitrust and other approval risks across countries.

• Financing Risk: Checks whether the buyer can pay, including debt availability, credit ratings and balance-sheet health.

• Precedent Mergers: Finds similar past deals and pulls their completion rates, timelines and risks.

• Timeline Events: Lays out past and upcoming milestones, such as filings, shareholder votes and court dates.

Then, a check

• Gap Analysis: Reviews everything collected so far for missing or overlooked information.

Finally, what’s next

• Catalyst Tracker: Flags upcoming events that could shift the deal’s odds, such as regulatory decisions, shareholder votes and financing deadlines.

Merger-Arb Specialists Review Agent Work

For the training deals, the researchers built example reports for the model to learn from. First, an AI agent wrote a post-mortem of each deal, covering how it unfolded and which early signals turned out to matter. It was the only part of the system allowed on the open web, since the outcome was already known. Then GPT-5.1 combined the agents’ research with the post-mortem to write a report as of the forecast date. It could cite only information public by that date, but it knew which facts deserved the most weight.

Veteran merger-arb specialists spent months reviewing the agents’ output on historical deals they had covered and flagging gaps. (The paper doesn’t say how many specialists were involved or give a more precise timeline, only that their feedback helped refine the agents’ responsibilities, prompts and output formats.)

The researchers also set the probabilities the model would learn from. Training on final outcomes alone makes forecasts overconfident, especially early in a deal, according to the paper.

They then fine-tuned GPT-4o on those reports and probabilities.

Finally, they tested it. For each test-set forecast date, the fine-tuned model read the agents’ research and produced five forecasts. The system used the median probabilities, while GPT-5 checked the five written reports against source documents and combined them into one.

Results

Most deals end well for shareholders. In the sample, about 14% collapsed without a better offer. So a model could look accurate just by predicting every deal would go through. To prevent that, the researchers gave each collapsed deal more weight in the score than each completed one.

The fine-tuned model beat every method they tested.

Read more

*I’m in New York for Cornell’s future of finance conference tomorrow. If you’re attending, it’d be great to meet up. Shoot me a reply or find me on LinkedIn.

Jane Street Locks Up More Computing Power

The Wall Street compute race rolls on.

Jane Street signed a five-year ~$13 billion contract with Crusoe for GPU clusters and infrastructure used to train and run AI models, Bloomberg reported last week. In April, the trading firm agreed to spend about $6 billion on CoreWeave’s AI cloud services, while separately investing $1 billion in CoreWeave. Jane Street has committed about $19 billion to AI computing this year.

Just as speed and colocation defined high-frequency trading in the 2010s, compute will define this era’s market makers. As Matthew Dixon, co-author of Machine Learning in Finance, told me in May:

This whole game is who’s got the most compute power. That ultimately determines who wins in this race. Everything else is kind of going to become secondary to that.

Ant Tests Self-Improving AI for Quant Research

Researchers from Ant Group, Princeton University and Stanford University released a paper last month on an AI system that generates trading ideas, tests them and uses the results to decide what to try next.

Called AQuA, the project has two separate AI research systems. One searches for trading factors in crypto markets. The other experiments with models that predict US stock returns over the next 30 minutes.

The “self-improving” part is fairly straightforward: Each experiment creates a record of what worked or failed, which the agents use to come up with the next idea. The underlying language model doesn’t retrain itself.

The headline result: Its US stock strategy produced a Sharpe ratio of 2.5 and posted positive results every year in a five-year backtest.

But the researchers were careful not to overstate what the backtest showed.

“No backtest, however strong, can by itself prove that a strategy is investable,” Gao Yunlong, an Ant Group researcher and one of the paper’s authors, said in an emailed response to questions from AI Street.

Gao said agents can run far more experiments than a human team, increasing the risk that they overfit the data or gradually turn a test set into part of the training process.

“The broader takeaway from our research is that finding signal may not be the hardest part. The harder problem is preserving causal validity, execution realism and institutional risk control as that signal moves from research into production.”

TL;DR

AI is making it easier to test investment ideas, shifting the human role toward judging whether the results make sense.

Man Group Cites AI in Quant Merger

As a Philosophy and English major (and failed coder), I always smile when I read lines like this:

Now, humanities majors interested in markets can do these jobs thanks to AI.

“The ability to think creatively will be a more relevant skill,” Taylor said.

That’s from a Business Insider story on Man Group’s decision to combine AHL and Numeric into Man Systematic, a $156 billion quantitative investing business.

AHL traditionally approached markets from the top down, with a focus on macro and trend following. Numeric was known for bottom-up equity investing. The businesses had separate code bases, and their researchers rarely collaborated.

Daniel Taylor, deputy chief investment officer of Man Systematic, said the technology has changed “on the margins” who can work as a quant. Coding and mathematics were once prerequisites. AI can now do more of the work required to turn an investment hypothesis into code that can be tested.

↑ This story at first sounds like these two were studying appellate law in the library basement in their free time until I read this line: (Emphasis mine.)

Grand and his prior lawyers parted ways after a federal appeals court ruled against him. When he decided to petition the Supreme Court on his own, he used artificial intelligence to learn the court’s intricate rules and study the work of top Supreme Court advocates.

“You really can’t copy the subject matter, but you can copy the formulation,” Grand said.

In markets, this points to a world where investors use AI to research more companies than they would have before, boosting volume. A topic we covered here:

ROUNDUP

What Else I’m Reading

• The AI Shift Turning Everyday Investors Into Mini Quant Funds | WSJ

• UBS demands new junior bankers show AI proficiency | FT

• Citadel Securities Urges SEC to Take Oversight of Some Wagers | Bloomberg

• Millennium Nears $100 Billion in New Era for Giant Hedge Funds | Bloomberg

• How AI can open new avenues for systematic investment research | T. Rowe Price

• Introducing Atum: A Foundation Model for Banking | Lloyds

• $24bn AI lab co-founded by an ex-Jane Street intern doesn’t like hiring quants as much as OpenAI | eFinancialCareers

Share AI Street

This Week in AI Street

Upgrade to paid

SPONSORSHIPS

Reach Wall Street’s AI Decision-Makers

AI Street reaches institutional investors, C-suite executives and Big Law attorneys at firms including JPMorgan, Citadel, BlackRock, Skadden, McKinsey, and more. Sponsorships are reserved for companies in AI, markets, and finance. Email sponsors@ai-street.co for more details.

CALENDAR

Upcoming AI + Finance Conferences

• Cornell Financial Engineering Manhattan 2026 Future of Finance & AI – September 11 • New York

  Hedge funds, banks and quants come together to explore how AI is changing finance, share practical lessons and exchange ideas on financial engineering.

• Machine Learning in Quantitative Finance – September 14–16 • New York

  Quant-focused conference on how financial institutions are applying machine learning across trading, portfolio construction, risk and investment workflows.

• AI for Finance Summit London – September 16 • London

  Invite-only summit on agentic AI in institutional finance, focused on how multi-agent systems are reshaping research workflows, PM coverage and alpha generation across hedge funds, asset managers and global banks.

• AI and Machine Learning in Quant Finance Conference – September 16 • Online

  CQF Institute event focused on AI and machine-learning advances in quantitative finance, with discussions geared toward practitioners and researchers.

• AI-Native Banking & Fintech Conference – September 29 • Salt Lake City

  Banking and fintech conference centered on agentic AI, with speakers from banks, fintechs and regulators discussing production use cases, risk, compliance and operations.

Thanks for reading!

I’m always happy to receive comments, questions, and feedback.

• Connect with me on LinkedIn, or

• Send an email to matt [at] ai-street.co

Manage how often you receive AI Street

Update Email Frequency

A big challenge in measuring how well AI predicts outcomes is look-ahead bias: quizzing a model on a test it's already taken.

AI providers generally don’t disclose their training data, so researchers are in the dark on whether a model has already seen the outcome they are asking it to predict.

But if the data has never been made public, you can be more confident that the model is actually making a prediction.

Duke has one such dataset. Its Fuqua School of Business has surveyed CFOs quarterly since 1996, with the Atlanta and Richmond Feds joining the project in 2020. The survey’s CFO Optimism Index has historically been a leading indicator of hiring and GDP growth, while research using the responses has also found that CFO expectations predict corporate investment.

The aggregate results are published. The individual executives’ answers remain confidential.

In a new paper, CFOs Meet LLMs, Duke finance professors John Graham and Campbell Harvey teamed up with Georgia State University professor Manish Jha to test whether an LLM could act as a digital twin for an individual executive.

The test: Could an LLM predict how a particular CFO answered in a particular quarter?

The researchers analyzed 6,075 survey responses from public-company CFOs between 2002 and 2025. In each response, the executive rated their optimism about the overall US economy on a scale from zero to 100.

GPT-5.4 was given the executive’s real name, company and ticker, along with details such as revenue, employment, foreign exposure and credit rating. It could search for company information, analyst reports and economic data available at the time, but was told to ignore anything published after the survey date. If a CFO had taken the survey before, the model also got up to 12 of their earlier confidential optimism scores.

Results

The researchers ran each prompt three times and averaged the results. The model gave very similar scores each time, with correlations of about 0.96 across the runs.

Compared with the real CFOs, the model was slightly more pessimistic and made the executives look more alike than they really were, producing fewer very high or very low scores.

Across all responses, the synthetic scores explained 27% of the differences in CFO optimism. The relationship held after the researchers accounted for each executive’s usual level of optimism, conditions during the quarter and the CFO’s most recent answer.

“It is not asking whether the model can separate persistent optimists from persistent pessimists,” Jha said in an email. “It is asking whether it can tell when a particular CFO turns more or less optimistic than their own usual level.”

Related Research

The researchers accounted for persistent differences among firms, each CFO’s most recent answer and whatever was moving everyone’s mood that quarter. The model’s score remained statistically significant.

But it worked much better when more history was available. Among responses with no previous answers, the model explained 10% of the variation. That rose to 33% when one to three previous answers were available and 49% when more than three were available. The model was much better at extending a known executive’s past views than building one from public information alone.

“The surprising part for me was how much the conditioning did, more than the raw capability of the model,” Jha said in an email.

The model isn’t a stand-in for a human CFO. It does not have access to the private information that makes an executive’s view valuable: current orders, internal costs, financing discussions and changes in corporate strategy. The authors described the synthetic scores as a possible supplement to human surveys, not a replacement.

The study also reconstructed past responses from 2002 through 2025 rather than making new predictions in real time. The researchers plan to test the system against future survey rounds.

For now, the system looks more useful for extending an existing panel of known respondents than for creating thousands of synthetic CFOs from scratch.

TL;DR

GPT-5.4 could predict how individual CFOs rated the economy, but it worked far better when given their previous survey scores.

Behind the paywall

Further commentary from Manish Jha on where digital twins are useful in finance — research and policy — and why he wouldn't trade on one yet.

Read more

Back in New York

I’m back in New York for a few days next week to attend Cornell’s Future of Finance & AI conference on Sept 11. It has a great agenda exploring many of the topics we’ve been covering at the vanguard of tech and markets. If you’re attending, let me know. It’d be great to meet in person!

Banks Ask Big Law to Share the AI Savings

I spent many years talking to litigators at big law firms when I covered white-collar crime for Bloomberg News. That’s where I came up close to the billable hour, which is straightforward enough, but strange when you see it in person. I saw many high-powered attorneys rushing back to the office because they were on the “clock.”

The FT reported this week that this is coming under pressure. Banks are asking law firms to cut their fees since AI is making routine work easier to complete. Top lawyers have long been paid on the backs of associates billing long hours, Morgan Stanley general counsel Eric Grossman told the newspaper. “Their compensation model is now extraordinarily unstable.”

Banks are still willing to pay top dollar for judgment, but not for rote work.

“We’re not yet seeing a formalised, firm-wide shift towards alternative fee arrangements, but it is likely to come,” said Gretta Rusanow, head of advisory services for Citi’s law firm group. She expects fixed fees for routine work, with the billable hour preserved for more complex matters.

I suspect the banks’ clients will eventually make the same argument.

If AI is making it easier to put deals together, why should the companies that hired the banks pay the same rates? With less deal friction, we’re likely to see more transactions done faster at lower fees.

Something similar happened when stock trading moved online in the 1990s. Technology made execution faster and dramatically cheaper, while the premium shifted from processing trades to providing advice and handling more complex transactions.

TL;DR

Banks want law firms to pass along their AI savings. Their own clients may soon demand the same.

Subscribe now

The Financial Models Built Like LLMs

What I find most interesting about AI is the surprising versatility of transformers, the architecture behind large language models.

Researchers and companies have trained transformer models on weather, grocery purchases, driving behavior, Netflix viewing data and bank transactions. In several cases, the models beat established benchmarks or existing production systems.

This is a topic we’ve covered quite a few times:

In a longer article in Chicago Booth Review that includes some of my previous coverage, I looked at what these models can do, where they still fall short and whether scaling them will be enough to compete with traditional quantitative methods.

It’s free to read:

The Risks AI Agents Pose to Central Banks

I was surprised, as I think most people are, to learn that no one knows exactly how LLMs work. Anthropic CEO Dario Amodei is fond of saying that AI is grown rather than built.

This uncertainty poses new regulatory questions as AI adoption picks up.

Markus Brunnermeier, an economist at Princeton University, thinks the technology could have widespread unintended consequences. In a paper presented to central bankers at Jackson Hole, he lays out a scenario in which AI makes markets less informative and more erratic, while giving market participants new ways to game central banks and regulators

AI agents, for instance, could learn from central-bank speeches, meeting minutes and past decisions until they become very good at predicting when policymakers will act. Brunnermeier argues that central banks might respond by communicating less clearly, relying on blunter rules or intervening more directly in markets.

TL;DR

If AI agents spread through financial markets, policymakers may struggle to understand them even as the agents learn to predict policymakers’ actions, according to a Princeton University economist.

ROUNDUP

What Else I’m Reading

• AI-driven cyber risk is top concern for global financial stability, watchdog says | Reuters

• JPMorgan curbed lending to Jane Street as trading firm muscled into bond market | FT

• NYSE Used Anthropic’s Project Glasswing to Find Cyber Flaws | Bloomberg

• A Glimpse Into Agentic Treasury Future | Treasury Today

• Anthropic Says New Fable 5.1 AI Model Is Cheaper, Better at Coding | Bloomberg

• Deutsche Bank Helps Shape Google Cloud’s New AI Solution for Financial Services | Deutsche Bank

• Building AI Systems for Capital Markets | Goldman Sachs

• Citadel Securities Notches Record $7.3 Billion Trading Haul | Bloomberg

• AI Futures: Terry Duffy, CME | Crain’s Chicago Business

• Terminal X Brings AI Operating System to Alternative Asset Managers with OpenAI | Terminal X

• XTX Markets’ top quants paid more than FTSE 100 CEOs | Financial News

• Agentic retail options trading triggers $1bn PFOF feeding frenzy | Global Trading

• Nvidia Agrees to Buy Hugging Face for $13 Billion | WSJ

Share AI Street

This Week in AI Street

RESEARCH

SPONSORSHIPS

Reach Wall Street’s AI Decision-Makers

AI Street reaches institutional investors, C-suite executives and Big Law attorneys at firms including JPMorgan, Citadel, BlackRock, Skadden, McKinsey, and more. Sponsorships are reserved for companies in AI, markets, and finance. Email sponsors@ai-street.co for more details.

CALENDAR

Upcoming AI + Finance Conferences

• 8th Annual Artificial Intelligence in Financial Services Conference – September 8–9 • London

  Senior financial-services event focused on real-world AI adoption across banking, insurance and fintech, including generative AI, risk management, fraud, compliance and governance.

• Cornell Financial Engineering Manhattan 2026 Future of Finance & AI – September 11 • New York

  Hedge funds, banks and quants come together to explore how AI is changing finance, share practical lessons and exchange ideas on financial engineering.

• Machine Learning in Quantitative Finance – September 14–16 • New York

  Quant-focused conference on how financial institutions are applying machine learning across trading, portfolio construction, risk and investment workflows.

• AI for Finance Summit London – September 16 • London

  Invite-only summit on agentic AI in institutional finance, focused on how multi-agent systems are reshaping research workflows, PM coverage and alpha generation across hedge funds, asset managers and global banks.

• AI and Machine Learning in Quant Finance Conference – September 16 • Online

  CQF Institute event focused on AI and machine-learning advances in quantitative finance, with discussions geared toward practitioners and researchers.

• AI-Native Banking & Fintech Conference – September 29 • Salt Lake City

  Banking and fintech conference centered on agentic AI, with speakers from banks, fintechs and regulators discussing production use cases, risk, compliance and operations.

Thanks for reading!

I’m always happy to receive comments, questions, and feedback.

• Connect with me on LinkedIn, or

• Send an email to matt [at] ai-street.co

Manage how often you receive AI Street

Update Email Frequency

As I’ve said a few times, AI is great at organizing messy data— the more tedious, the better. I’ve used local models on my Mac Mini to database ADV disclosures and patent applications and wrote about what I found:

You can also use AI to turn that cleaned-up data into an interactive dashboard—which is what I did with the H-1B filings.

These required Labor Department disclosures contain job titles, locations and wages for positions tied to H-1B sponsorship. I downloaded the data here and had OpenAI’s Codex standardize employer names, like “JPMorgan” and “J.P. Morgan,” identify duplicate case records and turn the data into a dashboard.

What the Filings Show

The dataset covers 141,941 unique filings from 91 finance firms, drawn from 26 quarterly releases spanning fiscal 2020 through fiscal 2026’s third quarter.

In the four quarters through June, H-1B filings from the 91 financial firms reviewed were roughly flat, declining about 0.9%. Nationwide filings fell 8.8% year over year in the first three quarters of fiscal 2026. (Nationwide H-1B-related applications peaked in fiscal 2022 and have remained below that mark since.)

Some important nuances: these filings are part of the H-1B sponsorship process, not evidence that someone was ultimately hired. One filing can cover multiple positions, and the salary figures show only the minimum base pay offered—not bonuses, equity or other compensation.

Three Firms Accounted for the Largest Increases

Fidelity, Citigroup and PayPal saw the biggest jumps in filings, adding 1,397 together. Excluding those firms, filings at the remaining firms declined about 8.2%.

The Banks Split

Citigroup and Morgan Stanley reported more filings. Bank of America, UBS and Deutsche Bank reported fewer.

(These are examples, not an exhaustive sector ranking. The dashboard and its download include every firm in the sample, including those with small changes or few filings.)

Payments Firms Moved in Opposite Directions

PayPal filed 23.1% more applications, while Visa and Mastercard filed 29.5% and 35.1% fewer, respectively.

AI Titles Only Make up 2%

Applications with explicit AI titles grew, but remained a sliver of the market, accounting for just 2% of filings, up from about 1% a year earlier.

JPMorgan and PayPal accounted for many of the explicit AI-title filings, with positions such as applied AI leads and machine-learning engineers. Bloomberg’s filings included a “Senior NLP Research Engineer.”

Meanwhile, 11 firms with at least 500 qualifying filings had no explicit AI titles in the archive. They include Goldman Sachs, Citigroup, Bank of America and Morgan Stanley. That does not mean those firms do no AI work: the data only captures job titles. Explore those firms.

What Firms Offer for AI-Titled Roles

Within employer-and-occupation groups containing both kinds of title, the raw median annualized offered wage floor was $170,000 for explicit AI titles and $145,000 for the comparison group—a $25,000 difference.

But much of that gap reflected differences in occupation, location and seniority. After accounting for those factors, the estimated difference narrowed considerably and should not be interpreted as a universal “AI premium.” See the fixed wage comparison and its limitations.

Data & Methodology

The analysis covers 141,941 H-1B filings linked to 91 selected finance firms. I standardized employer names and removed duplicate case records. Filings are not confirmed hires, and one filing can cover multiple positions. You can read the full methodology—including the employer-matching, AI-title and wage-analysis rules—on the dashboard.

Dashboard Access

Read more

I’ve previously written about the early uptake of AI trading agents, the broader shift from trading algorithms to trading agents, and how AI could push more trading into less-liquid securities.

More evidence of this trend came this week: German broker Scalable Capital opened its investment platform to ChatGPT, Claude and other AI assistants. Clients can use an AI agent to analyze their portfolios and prepare trades, which must be approved before execution.

Scalable has more than 1 million clients and more than €60 billion in assets on its platform. Co-CEO Erik Podzuweit called the service a “first step” and told Reuters that mass adoption was more likely once Scalable puts the capability directly inside its own app.

By my count, about a dozen brokerages let customers connect third-party AI assistants directly to their accounts, with permissions ranging from read-only portfolio access to autonomous trading.

• Alpaca: Executes paper and live trades

• Tradier: Constructs and submits trades

• TradeStation: Analyzes accounts and places trades after customer confirmation

• eToro: Trades autonomously within a dedicated portfolio

• moomoo: Backtests strategies and supports paper and live trading

• Public: Executes trades autonomously

• IG Australia: Provides read-only account access and cannot trade

• Robinhood: Trades autonomously through a dedicated account

• Interactive Brokers: Drafts orders for customers to review and submit

• Webull: Places and manages trades with an order preview

• Scalable Capital: Prepares trades for customer approval

Vanguard Bets $4 Billion on AI for Financial Advice

I think we’ll look back at the pre-AI era sort of like the way we look at the days of card catalogues now. For those of you too young to remember the smell of tens of thousands of aging index cards held in wooden cabinets, this is how you had to look up books before the internet.

This seems incredibly inefficient today—having to go to the library trying to find a book—but this was the best system available at the time.

As I’ve argued, AI cuts the tedium in our jobs by completing low-level administrative tasks. A 2019 Kitces.com survey of more than 1,000 advisers found that a typical lead adviser spent less than 20% of their time meeting existing clients and another 42% on back-office client work and administration.

Altruist tries to automate these processes. In February, the company unveiled an AI tool it said could analyze client tax documents and generate personalized strategies in minutes, helping trigger a selloff in wealth-management shares.

This week, Vanguard agreed to buy the startup for a reported $4 billion. The second-largest asset manager described its takeover of Altruist as an expansion of its financial-advice business through the acquisition of the “AI-forward wealth technology and custody platform.”

This is not solely driven by AI. Vanguard first invested in Altruist in 2020, and the acquisition also gives it a custody platform and closer ties to independent advisers.

TL;DR

AI should help lower the cost of financial advice by automating more manual tasks. Advisers could offset lower fees by serving more clients. I suspect other incumbents will follow Vanguard’s lead, using M&A to quickly boost AI adoption.

Subscribe now

What Else I’m Reading

• Young British investors trust AI more than TV or celebs | Finextra

• DBS moves AI-generated credit memos from a 150-person pilot to 1,500 bankers | DBS

• OpenBB Winds Down as a Company, Releases Entire Financial Technology Stack to Open Source | Rebellion Research

• Citi, HSBC, StanChart adopt Ant International’s forex AI tool | Reuters

• Harvard Is Selling a $699 Course Taught by A.I. Clones of Its Faculty | NYT

• Itoflow raises $2.5m for investment research AI agents | Finextra

• The GenAI premium: How alternative data is being repriced | Neudata

• ADIB appoints first chief AI officer | Finextra

• Stanley Druckenmiller’s AI writing has split Wall Street down the middle | BI

• PE Is Deploying an Army of AI Wonks to Embed in the Firms They Back | WSJ

• Revolut launches dedicated AI research lab built on its own foundation model | Finextra

• This startup sold Wall Street on AI. Its existential challenge is doing it again. | BI

• New Coinbase General Counsel Charts AI-Engineered Legal Path | Bloomberg Law

WorldQuant Says AI Is Boosting Solo Quant Research

AI makes it cheaper and faster to test ideas. This is a point I’ve heard often from folks building in the space. (I also discussed this in my interview with Tal Fishman at Lord Abbett.)

Here’s more evidence: WorldQuant, the quantitative asset manager founded and run by Igor Tulchinsky, holds an annual competition where students build and test predictive trading signals. This year, more than 150,000 people registered to compete individually in the contest, about double the 2025 figure. Multi-person teams increased much less, from over 4,000 to more than 5,300.

WorldQuant believes AI is helping explain the gap. Competitors used LLMs to brainstorm signals, generate variations on an idea, find overlooked data fields, test hypotheses and automate parts of the research process.

Fudan University finalist Ziming Xu used an AI-agent workflow to organize his research and avoid “brute-force testing,” according to comments provided by a WorldQuant spokesperson.

The 13 finalist teams will compete in Singapore on Oct. 6–7, developing and presenting new alphas to WorldQuant executives. The competition has a $100,000 prize pool.

Share AI Street

This Week in AI Street

Earlier this week, I wrote about how funds are trying to transfer institutional knowledge into AI models:

Business Insider published a related story:

Goldman’s engineers have a new challenge: turning AI agents into firm insiders

SPONSORSHIPS

Reach Wall Street’s AI Decision-Makers

AI Street reaches institutional investors, C-suite executives and Big Law attorneys at firms including JPMorgan, Citadel, BlackRock, Skadden, McKinsey, and more. Sponsorships are reserved for companies in AI, markets, and finance. Email sponsors@ai-street.co for more details.

CALENDAR

Upcoming AI + Finance Conferences

• 8th Annual Artificial Intelligence in Financial Services Conference – September 8–9 • London

  Senior financial-services event focused on real-world AI adoption across banking, insurance and fintech, including generative AI, risk management, fraud, compliance and governance.

• Cornell Financial Engineering Manhattan 2026 Future of Finance & AI – September 11 • New York

  Hedge funds, banks and quants come together to explore how AI is changing finance, share practical lessons and exchange ideas on financial engineering.

• Machine Learning in Quantitative Finance – September 14–16 • New York

  Quant-focused conference on how financial institutions are applying machine learning across trading, portfolio construction, risk and investment workflows.

• AI for Finance Summit London – September 16 • London

  Invite-only summit on agentic AI in institutional finance, focused on how multi-agent systems are reshaping research workflows, PM coverage and alpha generation across hedge funds, asset managers and global banks.

• AI and Machine Learning in Quant Finance Conference – September 16 • Online

  CQF Institute event focused on AI and machine-learning advances in quantitative finance, with discussions geared toward practitioners and researchers.

• AI-Native Banking & Fintech Conference – September 29 • Salt Lake City

  Banking and fintech conference centered on agentic AI, with speakers from banks, fintechs and regulators discussing production use cases, risk, compliance and operations.

Thanks for reading!

I’m always happy to receive comments, questions, and feedback.

• Connect with me on LinkedIn, or

• Send an email to matt [at] ai-street.co

Manage how often you receive AI Street

Update Email Frequency

Last month, we discussed how Bridgewater taught a model to handle research tasks more like one of its own investors:

…Bridgewater fine-tuned an open-weight model on examples labeled and reviewed by its investment experts, according to a statement this week from the hedge fund and Thinking Machines, the AI startup run by former OpenAI CTO Mira Murati. The goal was to teach the model what Bridgewater investors would consider relevant.

In other words, Bridgewater trained the model on its internal processes to give it institutional knowledge. And it’s not the only firm. A review of job postings at hedge funds, asset managers and private-equity firms shows others on Wall Street are trying to build similar firm-specific knowledge.

Firms are hiring AI specialists to codify how their investors work while retaining control of the data, workflows and feedback loops used to teach the AI. The firms still rely on frontier labs for broader use cases, as many of the partnerships we’ve covered show. But this goes beyond giving engineers access to coding tools.

Alex Gannes, founder of the investment management focused search firm, Banyak Group, described outside models and internal customization as complementary.

“A hedge fund is always positioning itself for an edge,” Gannes said. “Leveraging frontier labs and customizing AI solutions should ultimately support that edge.”

Investors are rightfully protective of that edge. And more broadly, companies are protective of their intellectual property and processes. This is a point Alex Karp, co-founder and CEO of Palantir, has been making in recent months.

“[Customers want] control over their compute, their models, their data stack and their alpha,” Karp told CNBC last month. “They want to know they own the means of production. It’s not being transferred to someone else.”

Karp is talking his book, but I tend to agree. Wall Street folks are a pretty paranoid bunch. I don’t see how firms ever get comfortable codifying how they work unless they know exactly where that knowledge lives and who controls it. (I’m not suggesting that frontier labs train—or would train—their models on customer data without permission. This is more of a cultural issue than a technical one.)

An AI-assisted review of more than a dozen current job postings at major hedge funds, asset managers and private-market firms shows firms pursuing three related goals: capture how their investment teams work, connect that knowledge to proprietary data and turn both into AI systems they control.

Related Coverage

The titles range from Fundamental Equities AI Engineer and AI Strategy Analyst to AI Credit Product Lead and Forward-Deployed AI Engineer. The use cases include idea sourcing, thesis monitoring, trade write-ups, portfolio construction, credit underwriting and investment-committee memos.

Behind those titles is a translation job: turning the judgment of portfolio managers, analysts and deal teams into prompts, retrieval systems, scoring frameworks and agents—and then determining whether the output is any good. The postings mention hallucination checks, audit trails, evaluating outputs without clean labels and getting direct feedback from investment teams.

Banyak Group’s Gannes thinks these teams could eventually support something broader than AI tools built for individual desks.

“A realistic goal in the hedge fund AI roadmap is to architect and scale an enterprise-level ‘hive mind’—linking autonomous AI agents into a single, coordinated cognitive network,” Gannes said. “Hiring a core team of applied AI engineers who can support every function within a firm would be a logical foundation for that effort.”

Behind the Paywall

Below, I break down about a dozen current postings from firms including Point72, D. E. Shaw, PIMCO, BlackRock, Ares and Carlyle. For each, I identified the target investment workflow and proprietary data involved. I then looked at how the firm’s knowledge would be encoded and the outputs tested. I included required investment experience and disclosed compensation where available.

Read more

Stripe, which provides the infrastructure for internet payments, agreed to buy OpenRouter, which provides infrastructure for using AI models.

OpenRouter evaluates AI requests and routes them to models based on price, speed, reliability and the complexity of the task.

I’ve heard repeatedly from financial firms that compute expenses are exploding as AI moves from isolated experiments into production. One hedge fund was spending about $24 million a year on LLMs. About one-third of that spending was wasteful.

A financial firm might use a smaller model to classify documents, a frontier model for complex research and a specialized system for extracting numbers. Prices and capabilities also change quickly. Without a routing layer, each application has to make those decisions itself—or default to an expensive model for everything.

I previously called the person overseeing this expense a “token master”: someone responsible for allocating compute budgets across departments and deciding which tasks warrant the most capable models.

OpenRouter tackles one part of the problem by routing each request to the model offering the best balance of capability, cost, speed and reliability.

Bridgewater Leaders Want to Tax AI Tokens

Hedge funders rarely call on lawmakers for more regulation. But Bridgewater’s Greg Jensen and Nir Bar Dea did just that in a New York Times opinion essay last Friday, urging a new tax on AI use and sweeping new regulations on the technology.

Bridgewater’s internal analysis estimates that AI could displace 18% of current U.S. jobs within five years. Jensen and Bar Dea warned that concentrating the benefits among companies and investors could provoke a political backlash.

The two proposed a tax on companies’ use of AI tokens, which they described as the closest equivalent to wages for machine labor. They argued that the current tax system makes replacing workers with AI comparatively attractive. That revenue could lower income taxes on workers and help the government acquire shares in leading AI companies.

They also called for strict safety standards covering models in development, publicly released systems and their use after deployment. The executives acknowledged that Bridgewater would be disproportionately affected by the policies they recommend.

Also in Bridgewater news:

Bridgewater Co-CIOs Back AI Startup for Stockpickers

Bridgewater co-CIOs Greg Jensen and Karen Karniol-Tambour have backed Multiplier, an AI startup founded by one of the hedge fund’s former investors.

The pair participated in a $6 million seed round led by Lux Capital. Y Combinator, Fortress Chairman Pete Briger and several of Multiplier’s hedge fund clients also invested.

Multiplier, previously called WithAI, was founded by former Bridgewater investment associate Ian McInnis, who worked under Jensen on large language model projects. Its coding agents filter news, make charts, monitor stocks and update investment memos. The platform connects those agents to a fund’s structured data, documents and internal research process.

Multiplier hosts the system on client infrastructure and interviews employees to teach the agents how each firm works. The company says it is live with six hedge funds and that users spend more than four hours a day on the platform.

CFTC Seeks Input on AI Compute Futures

The Commodity Futures Trading Commission is seeking public comment on proposed futures contracts tied to the computing power used to develop and run AI systems.

CME Group, Intercontinental Exchange and Architect Financial Technologies have announced plans to offer the contracts, subject to regulatory approval. They could allow AI developers and data-center operators to hedge changes in the cost and availability of compute, much as companies use energy derivatives to manage fuel costs.

Computing costs vary by chip, location, electricity supply and other infrastructure. The CFTC is asking whether the market needs common standards, including price indexes for settling contracts.

Compute is increasingly being treated as a scarce commodity as AI companies spend heavily on chips and data centers. A futures market could help them lock in prices, but only if the contracts reflect the capacity they actually use.

Google Wins Auction for Spirit Data to Improve AI

In simplified terms, AI scaling laws say more data + more computing power = better results from AI models.

The more computing power angle is well covered in the media. In May, I wrote:

Global tech companies have announced about $740 billion in AI-related spending for 2026, according to Morgan Stanley. That’s about what it cost to build the entire U.S. interstate highway system (in today’s dollars), which took 30+ years.

But the more data storyline is less covered. As we’ve written (and many others have covered) before pretty much all the publicly available data has been hoovered up. So much so that frontier AI labs are hiring experts, like ex-bankers and analysts, to create new data.

There are still vast stores of private data. Getting access is a slog. Labs need to establish who owns it, secure permission to release it, clear privacy and confidentiality checks, agree how it can be used and protected, and settle on a price.

That scarcity is creating a market for data left behind by failed companies. From Bloomberg Law:

Google Aims to Boost AI With Purchase of Spirit Airlines Data

Here’s the short of it:

Google won the bankruptcy auction with a $10 million bid for Spirit Airlines’ deidentified corporate data, including 100 million emails, 500 million Teams messages, billions of pricing and transaction records, and 30 million lines of code. Google says the cache will help improve its products and AI models. Mercor is the backup buyer after bidding $7.5 million.

TLDR

Corporate data that once had little value outside the company that created it is becoming an asset AI developers will bid for. Google’s $10 million bid for Spirit’s archive shows how far they will go to get it—and why investors should scrutinize the data behind an AI tool as closely as the model itself.

Share AI Street

ROUNDUP

What Else I’m Reading

• Hudson River Trading Posts Record $11.4 Billion Trading Revenue | BBG

• Jane Street Suffers Loss of About $15 Billion Following Troubles at Situational Awareness | WSJ

• Etched raises $700 million at $21 billion valuation | Quartz

• SpaceX Pitches Grok to Hedge Funds | Hedge Fund Alert

• AI Is Starting to Define Bank Internships as Much as Long Hours | BBG

• Optiver to take majority stake in power and gas trader Northpool | PR

Upgrade to paid

SPONSORSHIPS

Reach Wall Street’s AI Decision-Makers

AI Street reaches institutional investors, C-suite executives and Big Law attorneys at firms including JPMorgan, Citadel, BlackRock, Skadden, McKinsey, and more. Sponsorships are reserved for companies in AI, markets, and finance. Email sponsors@ai-street.co for more details.

CALENDAR

Upcoming AI + Finance Conferences

• 8th Annual Artificial Intelligence in Financial Services Conference – September 8–9 • London

  Senior financial-services event focused on real-world AI adoption across banking, insurance and fintech, including generative AI, risk management, fraud, compliance and governance.

• Cornell Financial Engineering Manhattan 2026 Future of Finance & AI – September 11 • New York

  Hedge funds, banks and quants come together to explore how AI is changing finance, share practical lessons and exchange ideas on financial engineering.

• Machine Learning in Quantitative Finance – September 14–16 • New York

  Quant-focused conference on how financial institutions are applying machine learning across trading, portfolio construction, risk and investment workflows.

• AI and Machine Learning in Quant Finance Conference – September 16 • Online

  CQF Institute event focused on AI and machine-learning advances in quantitative finance, with discussions geared toward practitioners and researchers.

• AI-Native Banking & Fintech Conference – September 29 • Salt Lake City

  Banking and fintech conference centered on agentic AI, with speakers from banks, fintechs and regulators discussing production use cases, risk, compliance and operations.

Thanks for reading!

I’m always happy to receive comments, questions, and feedback.

• Connect with me on LinkedIn, or

• Send an email to matt [at] ai-street.co

Manage how often you receive AI Street

Update Email Frequency

FUNDRAISING

Despite it being mid-August, lots of fundraising news this week:

UBS, HSBC Invest in Financial AI Startups

Over the past year and a half, banks have started buying stakes in the AI startups trying to automate their employees’ work.

This week, UBS backed Finster AI and HSBC invested in Model ML. FT Partners, which led Model ML’s $75 million Series A last year, is rolling out the platform firmwide. JPM invested in Rogo in April 2025, followed by Wells Fargo in January 2026.

These companies sit between the models and the bank. Their software connects to internal data, applies permissions and turns the output into something a banker can use: a presentation, financial model, research note or client briefing. Banks can swap the underlying model. Replacing the system wired into their data and workflows would be much harder.

Banks continue to build their own AI systems and sign deals with Google, OpenAI, Anthropic and other large technology providers.

TL;DR

The investments suggest banks are beginning to treat AI as an operating layer for financial work. The models underneath may be replaceable. The system that connects those models to the bank’s data, permissions and workflows may not be.

Quartr Raises $18M to Structure Company Data for AI

Quartr, a Stockholm-based provider of public-company research data, raised $18 million to expand its products and global coverage. Existing investor Altos Ventures led the round and became Quartr’s largest shareholder. The financing follows raises of $6 million in 2024 and $10 million last year.

The company collects earnings calls, transcripts, filings, presentations and other investor-relations material that remains scattered across corporate websites, conference lines and webcast providers. It links the audio, documents, speakers and company events in a common dataset, allowing analysts and AI systems to search across the material and trace results back to the original source.

Named API customers and partners include Perplexity, Yahoo Finance, Rogo, RavenPack and TradingView. The company says it now covers more than 15,000 public companies across over 65 markets and serves more than 800 financial institutions and technology companies.

ICYMI

The Financialization of “Compute”

Back in July 2025, I wrote:

New commodities often become financialized once volatility and market demand make risk hedging necessary.

That moment may be arriving for compute, with some well-known Wall Street traders believing compute demand will rival and then exceed the demand for oil.

Back in May, I highlighted this quote from DRW founder Don Wilson, who has a history of bringing new markets to the mainstream.

“The total dollars spent on compute will, over the next 10 years, exceed total dollars spent on oil.” DRW’s Don Wilson to the WSJ.

He invested in Silicon Data, a company that provides GPU pricing data and benchmarking services to hedge funds, banks and AI firms.

Just about a year later, Silicon Data raised $30.5 million in the initial closing of a Series A led by the Valor Atreides AI Fund. The round also drew backing from CME Group and a mix of trading firms, asset managers and technology investors. The New York-based company didn’t disclose its valuation.

CME plans to launch cash-settled futures tied to GPU rental costs, subject to regulatory approval, using Silicon Data’s benchmarks as the reference price.

Silicon Data publishes nine indices covering GPU rental prices and large-language-model costs, along with data on supply, utilization and pricing. The company says its platform has more than 1,000 registered users across semiconductor manufacturing, AI development and financial services.

Part of the new capital will fund SiliconMark, a system for measuring the performance of GPU infrastructure. Clusters built with the same chips can produce different results depending on their networking, configuration and the way their components are connected.

Silicon Data wants to standardize those performance differences, giving market participants a way to measure—and eventually hedge—the risk that a cluster delivers less computing power than expected. The company says the measurements could also support the physical delivery of computing capacity.

The remaining capital will go toward additional pricing benchmarks, institutional data products and infrastructure for derivatives, insurance and credit markets.

Silicon Data previously raised a $4.7 million seed round in March 2025.

AI at Two Sigma, DE Shaw and Susquehanna

AI leaders from some of Wall Street’s largest quantitative investment firms discussed how agentic AI is changing research, model evaluation, compute spending and hiring during a panel at Berkeley RDI’s Agentic AI Summit. The “Agentic AI in Capital Markets” speakers were:

• Jeff Wecker, chief technology officer at Two Sigma

• Jen Allum, senior vice president and co-head of generative AI at D.E. Shaw

• Ali Nazari, head of deep learning research at Susquehanna International Group

• Li Deng, chief AI officer at Vatic Investments and former chief AI officer at Citadel

“It’s becoming very clear, as we look at various metrics, that those who adopt sooner and faster are clearly becoming much more productive.”

—Jeff Wecker, Two Sigma, 09:21–09:31

“The hardest part didn’t disappear; it just moved. I found myself, instead of trying to generate ideas, trying to figure out which of the ideas it gave me were worth my time. These bottlenecks, these challenges, they don’t just disappear. They just move.”

— Ali Nazari, Susquehanna, 03:56–04:21

“Let lots of experiments run, be maximalist in the policy — build and buy and everything in between — and then see what catches fire and think about where we scale that, for whom and how.”

— Jen Allum, D.E. Shaw, 12:52–13:05

“We are on the eve of potentially an explosive use of compute demand… If behind those 1,800 employees there are a quarter of a million agents, all making their own compute demand, we really have to think about the implications and the return on investment.”

— Jeff Wecker, Two Sigma, 17:20–17:47

“If you want to bring that kind of system into the finance world, you are going to fail, because in financial markets, every time you put an order into the market, people react differently. It changes the environment much more than in the high-tech world.”

— Li Deng, Vatic Investments, 30:02–30:17

“Traditionally, we try something, we make a mistake, we learn from that mistake, and that’s how judgment is developed. Now that AI does most of the experimentation for us, the question is how that judgment is going to be developed.”

— Ali Nazari, Susquehanna, 20:42–21:01

ROUNDUP

What Else I’m Reading

• How AI is changing careers in electronic trading and HFT | eFinancialCareers

• Five Takeaways From Zuckerberg’s 6,500-Word Manifesto on AI | BBG

• Ex-Nubank CTO raises $85m for AI-native wealth advisory platform | Finextra

• Students created AI apps for the Norwegian Oil Fund | E24

• Prodigy Research: The Frontier AI Trading Research Lab | Y Combinator

• Some US adults are using AI for financial guidance but few trust it: Gallup | AP

• The new Wall Street career path | Business Insider

Share AI Street

This Week in AI Street

INTERVIEW + PODCAST

Upgrade to paid

SPONSORSHIPS

Reach Wall Street’s AI Decision-Makers

AI Street reaches institutional investors, C-suite executives and Big Law attorneys at firms including JPMorgan, Citadel, BlackRock, Skadden, McKinsey, and more. Sponsorships are reserved for companies in AI, markets, and finance. Email sponsors@ai-street.co for more details.

CALENDAR

Upcoming AI + Finance Conferences

• 8th Annual Artificial Intelligence in Financial Services Conference – September 8–9 • London

  Senior financial-services event focused on real-world AI adoption across banking, insurance and fintech, including generative AI, risk management, fraud, compliance and governance.

• Cornell Financial Engineering Manhattan 2026 Future of Finance & AI – September 11 • New York

  Hedge funds, banks and quants come together to explore how AI is changing finance, share practical lessons and exchange ideas on financial engineering.

• Machine Learning in Quantitative Finance – September 14–16 • New York

  Quant-focused conference on how financial institutions are applying machine learning across trading, portfolio construction, risk and investment workflows.

• AI and Machine Learning in Quant Finance Conference – September 16 • Online

  CQF Institute event focused on AI and machine-learning advances in quantitative finance, with discussions geared toward practitioners and researchers.

• AI-Native Banking & Fintech Conference – September 29 • Salt Lake City

  Banking and fintech conference centered on agentic AI, with speakers from banks, fintechs and regulators discussing production use cases, risk, compliance and operations.

Thanks for reading!

I’m always happy to receive comments, questions, and feedback.

• Connect with me on LinkedIn, or

• Send an email to matt [at] ai-street.co

Manage how often you receive AI Street

Update Email Frequency

Milos Maricic has a unique view into how the world’s largest asset owners are using AI.

Maricic and his team at SPEC Research work with sovereign wealth and pension funds as they figure out where the technology actually helps with investing.

Over the past six months, his team has also tried to answer a straightforward question: What are the world’s largest sovereign wealth and pension funds actually doing with AI?

They reviewed more than 1,000 public sources, incorporated information supplied by institutions and contacted the funds directly to assess 27 sovereign wealth funds and 26 pension funds. And today they released their SWF AI Maturity Index report.

The big takeaway:

The world’s largest sovereign wealth and pension funds are rapidly adopting AI, but none can show that it has made them money.

Or as CPP Investments CEO John Graham summed up in the report: “Will [AI] help us make faster decisions? Yes. Will it help us make better decisions? TBD.”

What Maricic and his team found was a wide gap between adoption and proof.

Singapore’s GIC has built a Virtual Investment Committee that includes an AI devil’s advocate trained on 44 years of internal deal data. The agent argues against an investment before the real committee sees it. CPP Investments has issued more than 2,100 Microsoft Copilot licenses across a firm of roughly 2,100 people, with 73% actively using them.

The strongest claim comes from Norway’s sovereign wealth fund, which says AI saves it about $100 million a year in trading costs. The fund has not published a methodology attributing those savings to the technology. More broadly, most funds do not track which parts of an investment decision came from AI and which came from people, making it difficult to tell whether AI improved a decision or simply made the process faster.

I spoke with Maricic about how sovereign wealth and pension funds are using AI. We talked about what sovereign wealth and pension funds are actually doing with the technology, why managers’ claims of AI-generated returns often fall apart under scrutiny, and why the growing volume of AI-generated research is creating another problem: the cost of verifying it.

I turned this interview, with the help of AI, into a new episode of my Alpha Intelligence Podcast. It has been dormant, but I plan to share more interviews going forward in this format, alongside the written transcripts.

Matt: How long have you been putting this report together?

Milos: About six months now. It is the first thing of its kind, so it has been ramping up really fast. I am just looking at how the response from the institutions has changed. When we launched the first one six months ago, it was like, “Do you want to comment on this?” And they were like, “No. Who are you? Just get out of here.”

Now it is more like, “Ah, yeah, sure. Let’s have you talk to our AI team. Let’s calibrate that.” So it has been getting some pickup.

ICYMI

Matt: So you are reaching out to these funds directly, saying, “Hey, we are trying to get a better sense of how you are thinking about this technology?”

Milos: Yeah. There are three things. First of all, the baseline job is to go through everything that is publicly available. The index has a database of more than 1,000 public sources.

Some of these institutions have internal information about themselves or their clients, and they allow us to use it specifically. Sometimes they do not, and that is fine. The third layer is that we reach out to them and say, “Hey, we are going public with this in a couple of weeks. Please help us understand what is going on.” More often than not, they come back, so we feel like we have a relatively robust picture of where the use is. Is it perfect? No. But this is a super-fast-moving field.

Matt: Right. And there are not many other resources looking at it like that.

Milos: That is the thing. Sovereigns are a niche to begin with, and even those who tackle that niche do not necessarily talk about AI use within it. So it is a niche within a niche.

Matt: What made you start it?

Milos: My last company was an AI-driven capital-allocation platform that got acquired by a sovereign.

So a lot of sovereigns and pensions were clients in the latter stages of that business. Business with sovereigns is a who-you-know business. It is very relationship-driven. We felt it was interesting for the team to continue in the same direction. We also felt we were doubling down on the advantage of being a little bit of insiders, knowing the right people, and so on.

Matt: I think of your role as somewhere between the vendors and all these solutions. If you are working at a sovereign wealth fund or you are a money manager, all this AI stuff can be too much.

So you are helping them sort out what is actually working, right?

Milos: Exactly. If it is overwhelming for you and me just trying to follow what is going on, it is even more overwhelming for them because they have all that capability. But what do you do with the capability?

Even if you have bought the story that AI is going to be big and you want to be part of it, what do you do? Where do you deploy? Do you deploy in the biggest, most inflated-valuation companies out there? Do you look for the nuggets? Do you go through managers?

If we focus on the manager side, a lot of what allocators do is manager selection. There is an explosion of managers coming to them and saying, “We use AI. Look at our backtests. Look at our track record. It is all amazing.” Allocators are prudent about it, and they have every reason to be, because a lot of this is just bogus.

Some of it is interesting, but you have to distinguish between the two. One of the things we produce is a framework called the SPEC test that we coach allocators to use with managers. It breaks down their AI claims in a structured way to understand what holds water and what does not.

One priority for the second half of this year is working with the Institute for Sovereign Investors to take the SPEC framework and turn it into a more industry-wide due-diligence framework. It is sorely needed. Right now, what is happening within allocators around AI claims is completely ad hoc.

Matt: What are fund managers saying they are doing with AI?

Milos: The most typical thing is: “We backtested this, it worked amazingly well, and our Sharpe ratio ended up being three.” Then you look into the backtest and realize the model had a little glimpse of the future. It actually saw where the market was going.

Another thing is riding heavily on the track record of the managers themselves. But there is loads of research showing that, if you use AI to do fully automated investing, the aspects that make up a manager’s track record are what AI can replicate most easily. The more idiosyncratic parts are more difficult to pick up.

When we make managers go through the framework, we ask them to decompose their claims. We probe them. We say: “Tell me about a time when the model was wrong. What happened? How did you adapt it? How did you fix it going forward?” Or: “Tell me about a time when it was overruled by a human.” That gives you a window into their internal process, so you can understand where the boundary is between human capacity, human oversight, and AI. The framework has about 14 questions that we drill into.

Matt: A sovereign wealth fund will call you up and say, “XYZ reached out. I want you to take a look. Do you know about them? Maybe you already do.”

Milos: Yeah, exactly.

Matt: What are some red flags you have seen?

Read more

I’m traveling to Rhode Island to visit friends and family for a couple weeks (if you’re in the area, reach out!). I plan to keep AI Street publishing as usual while stateside.

Bots now generate more web traffic than humans. Cloudflare had expected to reach this milestone next year, but it happened in June.

While humans have limited time and attention, agents can theoretically run 24/7 and juggle multiple webpages at once.

Take this dynamic into markets and you’re likely going to see a lot more trading. We discussed this trend earlier this year: when Italy banned ChatGPT for a month in 2023, investors narrowed their trading to well-known companies, and once ChatGPT came back, trading broadened out again.

And in June, I wrote:

I’m having a hard time seeing trading not marching higher and permeating into more markets given current AI trading trends. It didn’t make a lot of economic sense to spend limited human time researching smaller markets. Now it does.

Continuing this trend, here’s Bloomberg with a story on how retail traders are leveraging AI for more sophisticated trading strategies.

The article highlights a couple points we’ve covered here along with nice anecdotes.

1. AI is (obviously) not a magic money-making machine.

2. Crowding is a real risk.

   1. The article quotes HFT Irene Aldridge, who I interviewed here, saying “AI models are parsing the same data, using very similar models and outputting very similar results to everyone,” she said. “We’re going to see a lot more volatility as a result of this.”

3. AI could amplify losses.

   1. The article quotes AI and finance researcher Alicia Vidler, who I also interviewed, on the risks: “Lots of trading systems are not designed for volatile environments,” she said. “You have much more risk in a down market.”

4. False AI confidence needs to be mitigated.

   1. “It will answer incredibly confidently on anything you ask regardless if it knows it or not,” Joel Rieger, a former software-sales executive, told Bloomberg.

   2. This is a topic I’ve covered here: AI Replicates Human Investor Biases.

All that said, the rollout is moving quickly. AI-agent trading may be a novelty today, but I suspect it will become a standard offering.

AI Agent Trading

Since March, four retail brokers have launched or begun rolling out AI-linked trading tools, ranging from autonomous execution to AI-assisted order drafting.

• Public.com launched native agents that can execute user-approved workflows across stocks, options and crypto.

• Moomoo released API Skills, connecting outside agents to its market-data and trading infrastructure.

• Robinhood introduced dedicated accounts where third-party agents may execute trades without direct approval of each transaction.

• Interactive Brokers let customers connect Claude and other MCP-compatible AI tools, although clients must review and submit every order.

In about two months, more than 100,000 people had connected an agent and opened a dedicated Agentic account at Robinhood, CEO Vlad Tenev said on an earnings call last week. Robinhood said assets and trading volumes were also growing, though it did not disclose either figure.

In April, Moomoo, the trading platform owned by Futu Holdings, launched tools that let customers connect AI agents to analyze markets and prepare trades. Since then, the number of traders using its US API has more than doubled. (The figure includes all API users, not only those trading with AI agents, so it is a proxy for adoption.)

“It used to require meaningful coding chops to access trading APIs and do real volumes,” David Rodríguez, product lead at Moomoo US, told me. “The learning curve has shifted significantly, and we’re seeing the results unfold in real time.”

ROUNDUP

What Else I’m Reading

• Millennium Partners With Anthropic to Develop AI Risk Analyst | BBG

• JPM CEO Dimon leads new cross-industry effort to tackle AI risks | Reuters

• AI Skills More Valuable Than MBAs, 86% of Finance Executives Say | BBG

• The Investing Heavyweights That Backed Situational Awareness Before It Blew Up | WSJ

• Google shakes up AI leadership as DeepMind chief shifts role | Reuters

• Four Top Google A.I. Researchers Form New Start-Up | NYT

• Rabobank to invest €2 billion in building IT framework for AI | Finextra

• Cloudflare gives AI agents an identity and a wallet | Finextra

• Wall Street hit by wave of AI-powered ’vishing’ cyberattacks | Yahoo

• Goldman Sachs asset arm forms AI investing platform, memo shows | Reuters

• Xi Chen joins JPM as MD, Head of AI for Electronic Markets | LinkedIn

Share AI Street

This Week in AI Street

Upgrade to paid

SPONSORSHIPS

Reach Wall Street’s AI Decision-Makers

AI Street reaches institutional investors, C-suite executives and Big Law attorneys at firms including JPMorgan, Citadel, BlackRock, Skadden, McKinsey, and more. Sponsorships are reserved for companies in AI, markets, and finance. Email sponsors@ai-street.co for more details.

CALENDAR

Upcoming AI + Finance Conferences

• 8th Annual Artificial Intelligence in Financial Services Conference – September 8–9 • London

  Senior financial-services event focused on real-world AI adoption across banking, insurance and fintech, including generative AI, risk management, fraud, compliance and governance.

• Cornell Financial Engineering Manhattan 2026 Future of Finance & AI – September 11 • New York

  Hedge funds, banks and quants come together to explore how AI is changing finance, share practical lessons and exchange ideas on financial engineering.

• Machine Learning in Quantitative Finance – September 14–16 • New York

  Quant-focused conference on how financial institutions are applying machine learning across trading, portfolio construction, risk and investment workflows.

• AI and Machine Learning in Quant Finance Conference – September 16 • Online

  CQF Institute event focused on AI and machine-learning advances in quantitative finance, with discussions geared toward practitioners and researchers.

• AI-Native Banking & Fintech Conference – September 29 • Salt Lake City

  Banking and fintech conference centered on agentic AI, with speakers from banks, fintechs and regulators discussing production use cases, risk, compliance and operations.

Thanks for reading!

I’m always happy to receive comments, questions, and feedback.

• Connect with me on LinkedIn, or

• Send an email to matt [at] ai-street.co

Manage how often you receive AI Street

Update Email Frequency

“When will machines write the rules, not just support the rules?” Jensen recalled asking himself in the early 2010s. “When will their intuition actually be better than human intuition?”

At the time, the technology could do “great pattern matching,” Jensen said, but “there was nothing like intuition.” The pieces he believed an artificial investor would need—reasoning, ways to diagnose its conclusions, and methods for working with limited data—had not yet come together.

So Jensen went looking for them.

That search led him to OpenAI, then a little-known nonprofit that would become the face of the AI boom, where he became an early investor around 2016. Later, after a group of researchers he knew left OpenAI to start a new company called Anthropic, he wrote the company’s first outside check.

By late 2022, the year ChatGPT went mainstream, Jensen thought the missing pieces were finally coming together. The following year, Bridgewater created AIA Labs, short for Artificial Investment Associate, a separate research and investment unit tasked with building what the firm calls an “artificial investor.” It started with six people.

Bridgewater kept the group apart from Pure Alpha, its flagship macro strategy. The separation was intended to give the team room to rebuild the investment process around AI—and to keep the new systems from simply reproducing the views Bridgewater’s investors had already developed.

“I don’t want a copy of us,” Jensen said. “We’re so flawed. We’re so bad at so many things. I want something better than us.”

The assignment was broader than finding a new way to analyze data. AIA was supposed to take on more of the work an investor performs: gathering information, developing an explanation for what was happening in an economy or market, testing that explanation and turning it into an investment view.

AIA was not built around a single model or trading strategy, according to a person familiar with the effort. Instead, it combines Bridgewater’s causal time-series models with specialized and frontier language models, along with systems designed to test and diagnose their outputs. The broader goal is to generate insights across markets, economies and trading strategies.

Bridgewater tested the strategy with its own money first, beginning with roughly $100 million inside Pure Alpha in late 2023. The firm opened it to outside investors in July 2024, attracting almost $2 billion.

In March 2025, Bridgewater CEO Nir Bar Dea said the fund was producing “unique alpha that is uncorrelated to what our humans do.”

Today, AIA is no longer a six-person experiment. The unit has grown to more than 75 scientists, engineers and investors, according to the person familiar with the effort. The strategies now manage about $4.5 billion. AIA Macro has returned 11.3% annualized since inception, with returns that have shown low correlation with markets and other managers, including Bridgewater’s Pure Alpha.

Bridgewater has started using AIA’s systems inside Pure Alpha to accelerate research and systematize investment ideas, the person said.

A Bridgewater spokesperson declined to comment on the hedge fund’s AI efforts.

Read on for how Bridgewater went from writing investment rules by hand to letting an artificial investor manage billions.

Read more

AI and Alpha: Why Technology Alone Won’t Be Enough

Andrew Chin, chief AI officer at AllianceBernstein, published a piece this week that goes through the history of technology in the investment process.

Chin highlights how spreadsheets, the web and alternative data each gave early adopters an edge before eventually becoming standard tools available to everyone. AI is on the same path.

Eventually, everyone will have access to roughly similar models. What separates firms will be the judgment they bring to them: what data they focus on, what questions they ask and the viewpoints they build into the technology. We highlighted this earlier this month when Bridgewater trained an AI model to think like one of its investors.

AI is an amplifier. If you’re bright, these tools make you that much more effective. You’re able to scale what you already do well. If you’re not so bright, or maybe just a bit careless, well, that scales too.

SPONSOR

The earnings headline was in every feed: options had priced a $190 billion swing on the print.

Microsoft asking Amazon and Google for cloud capacity help, four law firms circling with shareholder-suit deadlines, and the launch of its new cyber-defense AI models were in the trade and legal press; the positioning signal, scattered across outlets no single feed carries.

Seltz returns all of it, raw. Full articles, first-person reporting, within the hour of publication. No summaries, no sentiment scores. Your models do the distillation better than any vendor, on input nothing has influenced.

Search any name across any day in history the same way. That’s how teams build and test a signal before they trade it live. Backed by B Capital and Speedinvest.

Want the full picture on every name you trade? Sign up for a demo — bring 10 questions your desk actually asks, and we’ll run them live.

JPM Moves AI Into Its Core Tech Stack

Last month, I wrote that responsibility for AI might eventually be absorbed into the CTO or CIO role. That appears to be happening at JPMorgan.

The bank is placing the office that led its AI push inside Global Technology, signaling that AI is becoming core technology rather than a standalone initiative. CTO Scot Baldry is taking over the firmwide AI program as Teresa Heitsenrether prepares to retire at year-end, according to Business Insider. The central team will continue setting strategy and governance, while individual businesses will increasingly be responsible for deploying AI and proving its value.

In other words, AI is becoming too central to remain siloed within a dedicated division. I suspect this shift from side project to core function will happen across companies as AI becomes more embedded in workflows.

Asset Managers’ AI Spending Surges

Asset managers are pouring money into AI without much agreement on how much is enough.

Clearwater Analytics commissioned a survey of 178 senior executives at hedge funds, private-markets firms, insurers and traditional asset managers across the US, Europe and Asia-Pacific.

Sixty-three percent said their firms increased AI spending by more than 50% during the past year. None reported a decrease. Yet 66% worried they were overinvesting, while 25% thought they were not spending enough.

AI is also moving closer to the portfolio. Forty-three percent said it is involved in between 25% and 49% of their investment decisions. Thirty-eight percent reported similar adoption in risk management and 34% in operations.

The survey does not identify the firms, their assets or what qualifies as AI involvement. It also combines generative AI with older machine-learning and automation systems.

ROUNDUP

What Else I’m Reading

• ICE puts fixed income pricing data on leading AI platforms | Finextra

• OCBC Uses AI To Ease Wealthy Clients’ Onboarding Pain | WealthBriefingAsia

• HSBC to open Singapore AI centre of excellence | Finextra

• The Rise of Million-Dollar Companies With Just One Employee | WSJ

• Situational Awareness Seeks to Raise Capital After AI Rout | FT

• Bloomberg to Buy Private Markets Data Platform Canoe Intelligence | Yahoo

• People Building AI Want Governments to Slow Down the AI Race | Euronews

• Citadel spin-offs are favorites for hedge funds investing externally | BI

• How Bridgewater Built an AI Analyst That Does Hours of Expert Research in Minutes | YouTube

• CME’s Debut of 24/7 Gold Futures Trading Met With Strong Demand | BBG

• Flow Traders Selects CoreWeave for Model Training | Press Release

• BNY to adopt blockchain technology for records in new era for Wall Street | FT

Share AI Street

This Week in AI Street

Thanks for those of you who joined yesterday for AI Street’s first Substack Live. It was a great chat with Ex-Millennium quant Arman Khaledian on the latest on AI adoption, changing research workflows and Wall Street’s growing compute arms race.

If you’d like to suggest a guest or a topic, please reply to this email or reach out at matt [at] ai-street.co.

Upgrade to paid

CALENDAR

Upcoming AI + Finance Conferences

• Ai4 2026 – August 4–6 • Las Vegas

  Large enterprise AI conference with a finance track and broad coverage of applied AI, agentic systems, infrastructure, governance and deployment lessons.

• 8th Annual Artificial Intelligence in Financial Services Conference – September 8–9 • London

  Senior financial-services event focused on real-world AI adoption across banking, insurance and fintech, including generative AI, risk management, fraud, compliance and governance.

• Cornell Financial Engineering Manhattan 2026 Future of Finance & AI – September 11 • New York

  Hedge funds, banks and quants come together to explore how AI is changing finance, share practical lessons and exchange ideas on financial engineering.

• Machine Learning in Quantitative Finance – September 14–16 • New York

  Quant-focused conference on how financial institutions are applying machine learning across trading, portfolio construction, risk and investment workflows.

• AI and Machine Learning in Quant Finance Conference – September 16 • Online

  CQF Institute event focused on AI and machine-learning advances in quantitative finance, with discussions geared toward practitioners and researchers.

• AI-Native Banking & Fintech Conference – September 29 • Salt Lake City

  Banking and fintech conference centered on agentic AI, with speakers from banks, fintechs and regulators discussing production use cases, risk, compliance and operations.

Thanks for reading!

I’m always happy to receive comments, questions, and feedback.

• Connect with me on LinkedIn, or

• Send an email to matt [at] ai-street.co

Manage how often you receive AI Street

Update Email Frequency