Daily briefing

AI in share trading: what matters this week

Independent, technically-grounded commentary on artificial intelligence in equity trading. Curated daily from primary sources across regulators, exchanges, banks, brokers, funds and academic research. No hype, no financial advice - just what a serious builder or portfolio manager should be tracking.


Sunday, 25 October 2026

Regulation

FCA chief says threats to market integrity are 'as serious as I have seen them', and names agentic coordination as a manipulation question

In a 22 October Mansion House speech, FCA chief executive Nikhil Rathi pointed to autonomous AI agents, 24/7 trading and new forms of market abuse, and said staff judgement will be needed on when agentic coordination becomes market manipulation.

Source: Financial Conduct Authority (speech by Nikhil Rathi, 'Market integrity in a digital age', City Dinner, Mansion House, 22 October 2026)

Saturday, 24 October 2026

Friday, 23 October 2026

Thursday, 22 October 2026

Regulation

FCA publishes good and poor practice on AI, including governance layers and an inventory for AI agents

On 19 October 2026 the FCA published illustrative examples of good practice and areas for improvement in firms' use of AI. It says the examples create no new requirements, and it singles out agentic AI as likely to need stronger controls. The page has no trading-specific guidance.

Source: Financial Conduct Authority (AI in financial services: good practice and areas for improvement)

Wednesday, 21 October 2026

Regulation

CFTC publishes the agenda for its first Frontier Forum on super intelligence and markets, set for 28 October

On 20 October 2026 the CFTC's Innovation Task Force announced the agenda and panelists for the inaugural Frontier Forum, a public roundtable on how advanced AI is being used at the frontier and what it means for U.S. markets. The series announcement in September described the first forum as focused on artificial intelligence and agentic finance. No rules or proposals were announced.

Source: U.S. Commodity Futures Trading Commission (Innovation Task Force, release 9312-26)

Tuesday, 20 October 2026

Regulation

FCA says it will build a trusted framework for agentic finance: agent identity, delegated authority, accountability and execution

In a speech on 19 October 2026 the FCA's chief data, information and intelligence officer set out its response to the Mills Review. The regulator says it is embedding six of the review's seven recommendations into live work, and will work with industry, other regulators and frontier AI labs on a framework for agents that act for consumers, including on investments. No new trading rules were announced and no timeline was given.

Source: Financial Conduct Authority (speech by Jessica Rusu, Bloomberg UK Policy Series)

Friday, 16 October 2026

Regulation

Taiwan's FSC consults on agentic AI security for financial firms: scenario-by-scenario governance, a three-axis risk rating and agents treated as non-human identities

On 13 October 2026 Taiwan's Financial Supervisory Commission issued a consultation document on how financial institutions should secure agentic AI. It proposes managing each agent by application scenario, rating risk on data sensitivity, impact of actions and capability boundaries, and applying zero-trust controls to the agent as a non-human identity. Comments are open for 30 days.

Source: Taiwan Financial Supervisory Commission

Thursday, 15 October 2026

Regulation

The FSB's final AI sound practices add agentic AI considerations: minimum permissions, kill switches and human-in-command oversight

The Financial Stability Board published its final report on Sound Practices for the Responsible Adoption of AI on 15 October 2026. It sets out 12 practices, says the report now includes additional considerations for agentic AI and multi-agent systems, and states there is little evidence at present of GenAI being used for fully autonomous trading.

Source: Financial Stability Board

Tuesday, 13 October 2026

Regulation

The FSB has finalised its sound practices for AI in finance and asks the G20 to champion them, as a toolkit rather than a rulebook

The Financial Stability Board's Chair's letter to G20 finance ministers on 12 October listed the finalised Sound Practices for the Responsible Adoption of AI among its deliverables, after the June consultation; the final report itself was published on 15 October. The release describes them as adaptable to a firm's size and level of AI use and does not say what changed.

Source: Financial Stability Board

Saturday, 10 October 2026

Friday, 9 October 2026

Thursday, 8 October 2026

Wednesday, 7 October 2026

Tuesday, 6 October 2026

Monday, 5 October 2026

Wednesday, 30 September 2026

Tuesday, 29 September 2026

Saturday, 26 September 2026

Friday, 25 September 2026

Wednesday, 23 September 2026

Market-structure

A consumer AI agent, not a trading bot, just moved the market: Meta's Muse wiped billions off banks, insurers and travel stocks in one session

Meta's Muse hit No. 1 on Apple's US App Store days after launch, and a Goldman Sachs trading desk note warning it threatens industries built on customer passivity sent the S&P 500 Financials Index down 2.4%, with Charles Schwab off 6.7%, Planet Fitness down 11% and Booking Holdings and Expedia both down close to 4%.

Source: Investing.com (via Yahoo Finance)

Tuesday, 22 September 2026

Monday, 21 September 2026

Research

A systematic audit of 15 academic trading-agent papers finds every one carries a security hole and most fail basic market stress

FARSIGHT, a new evaluation framework submitted 17 September 2026, tests published financial LLM trading schemes against flash-crash-like turbulence and three attack categories; 80% fail at least one core robustness metric and 100% exhibit at least one security vulnerability, and the paper shows the same failure can trigger a crash by accident or be forced by an adversary with minimal resources.

Source: arXiv

Sunday, 20 September 2026

Saturday, 19 September 2026

Regulation

SEC grants a five-year exemption that lets tokenized-stock venues run 24/7, permissioned-AMM trading outside existing exchange rules

Chair Paul Atkins' Innovation Exemption, announced 17 September, gives Tokenized Securities Venues and their liquidity providers five years' relief from standard exchange and dealer-registration rules, provided holders keep the same dividend and voting rights as ordinary shareholders and issuers get a right to object before a digital version of their stock trades.

Source: CNBC

Friday, 18 September 2026

Thursday, 17 September 2026

Wednesday, 16 September 2026

Tuesday, 15 September 2026

Monday, 14 September 2026

Sunday, 13 September 2026

Saturday, 12 September 2026

Friday, 11 September 2026

Thursday, 10 September 2026

Wednesday, 9 September 2026

Regulation

The FCA's frontier AI review sets no new trading rules, but it tells firms the thing that gets audited is the harness around the model, not the model itself

A 2 September multi-firm review finds frontier AI models are exposing existing weaknesses in firms' vulnerability management, access controls and remediation capacity faster than those processes can keep up, and that the value a firm gets from a frontier model depends less on which model it runs than on the governance, tooling and human oversight wrapped around it.

Source: Financial Conduct Authority

Monday, 7 September 2026

Market-structure

Anthropic's Fable 5.1 didn't cut its sticker price, it cut the cost of an agent staying in context all day, and that is the number that decides whether continuous market-monitoring agents are economical

Fable 5.1 keeps Anthropic's list price at $10/$50 per million input/output tokens but drops cached-token reads from $1 to $0.25 per million, a 75% cut Anthropic says makes highly agentic workloads roughly 45% cheaper. For a trading agent that has to hold order books, news and position state in context for hours, that is the line item that mattered.

Source: LLM Stats

Sunday, 6 September 2026

Infrastructure

Tokenized markets expose a gap this site's architecture coverage has flagged before: an AI trading agent can act on an asset only as fast as the rules governing it become machine-readable

Brickken co-founder Edwin Mata argues the bottleneck in agentic capital markets isn't the AI or the tokenized asset, both already exist, but the layer between them: ownership, permissions and compliance still live in scattered databases and legal PDFs an autonomous agent cannot parse or enforce.

Source: Hedge Think

Saturday, 5 September 2026

Regulation

A House Financial Services hearing put NYSE and Circle in the room to answer a question this site keeps returning to: who is an autonomous trading agent, legally, when it talks to another agent

At a September 2 hearing, Rep. Bill Foster framed agent-to-agent finance as starting with "who the heck are you" before it can start with a trade, while Circle's Heath Tarbert told the committee the always-on, machine-readable settlement rails agentic trading needs are still in "early innings." NYSE's Lynn Martin was in the room too, putting exchange infrastructure on the same table as stablecoin plumbing.

Source: PYMNTS

Friday, 4 September 2026

Regulation

A Husch Blackwell analysis names the specific statutory gap agentic trading opens under the Commodity Exchange Act: intent-based rules built for a mind, not a model

"Agents at the Gate," by Husch Blackwell's Jeff Le Riche, Sarah Razaq Sallis, Kip Randall and Bhargavi Kalaga, walks past the familiar broker-definition and investment-adviser questions to a narrower and harder one: the CEA's anti-spoofing and manipulation provisions require proof of intent, and nobody has settled whose intent an autonomous agent's trade reflects.

Source: Government Enforcement, Compliance & Investigations Report (Husch Blackwell LLP)

Thursday, 3 September 2026

Wednesday, 2 September 2026

Tuesday, 1 September 2026

Saturday, 29 August 2026

Market-structure

Crypto exchange KTX launches a Skills Kit connecting Claude, ChatGPT, Codex and Cursor directly to live market data, portfolio management and order execution

The August 18 release gives agents access to real-time prices, order books and KTX's own AI-generated market intelligence, then lets the same agent act on it, joining a wave of exchange-side agent tooling that includes Coinbase's MCP server, OKX's 163-tool Agent Trade Kit, Kraken's agentic CLI and Binance's AI Agent Skills.

Source: Decrypt (Chainwire press release)

Friday, 28 August 2026

Regulation

A Cox Media Group investor-protection feature asks what happens to a retail customer's money when an AI trading agent errs, and finds SIPC coverage was never built to answer that question

The piece uses OpenAI developer Nik Pash's 'Lobstar Wilde' trading bot, which misread a plain-language post in February and sent its entire crypto treasury, roughly $250,000, to a stranger instead of a small tip, as the cautionary example for a wider point: SIPC insures against a broker's failure, not against an autonomous agent executing a trade a customer never intended. The article notes Robinhood, Public, SoFi, Coinbase and Webull have all rolled out AI trading agents to retail customers this year, and that regulators and Congress are now asking how investors are protected when the agent, not the customer, makes the mistake.

Source: Cox Media Group

Thursday, 27 August 2026

Research

Elm Wealth's Crystal Ball Challenge gave Claude, ChatGPT, Gemini and Grok real historical Wall Street Journal front pages and withheld the actual market outcomes, finding Claude beat human participants in 76 percent of roughly 200 sessions

ChatGPT beat human participants in 63 percent of sessions in the same study, but researchers found all four models systematically took on more risk relative to the information they were given than the trade context justified. The study is one of the more rigorous public attempts to isolate how large language models actually behave when asked to trade on real historical information rather than a live, backtested feed.

Source: Elm Wealth

Institutional

A Wall Street Journal analysis finds nine major technology companies carrying roughly $3 trillion in off-balance-sheet AI infrastructure commitments, with Morgan Stanley warning the structure makes true leverage hard for investors to see

About $1.2 trillion of the total comes from unstarted leases and $1.9 trillion from purchase commitments tied mostly to AI data-centre buildout, none of which shows up on a standard balance sheet. Morgan Stanley separately projects total AI infrastructure spending will reach $3.2 trillion by 2028, with roughly $1.75 trillion of that needing to be raised through credit markets.

Source: TipRanks

Retail

Munich-based Scalable Capital becomes the first European bank to let retail customers connect ChatGPT, Claude or Grok directly to their brokerage accounts and place trades by prompt

The new Agentic Investing feature lets more than 1 million customers across roughly EUR 60 billion in assets execute trades, manage watchlists and set price alerts through a connected AI assistant, with two-factor authentication and manual trade approval required before execution. The launch follows a separate research finding that Claude beat human traders in 76 percent of sessions in a historical-data backtest, though researchers noted the AI systems generally took on more risk than the trade context justified.

Source: Fortune

Wednesday, 26 August 2026

Research

Google Cloud's 2026 State of AI Infrastructure report names "agent sprawl" as the defining operational risk of scaling agentic AI, with 79% of tech leaders citing security, governance or operations as their biggest blocker

The report frames the core problem in terms this site's trading-desk readers will recognise immediately: managing thousands of autonomous agents scattered across platforms without losing visibility or control, the same challenge already playing out on exchange-side agentic trading rollouts this site has tracked all month.

Source: Google Cloud Blog

Market-structure

Nvidia reports fiscal Q2 results today with analysts pricing in $91.8bn in quarterly revenue, and desks are treating the print as the decisive test of whether the AI trade can hold after August's chip selloff

Chip stocks were mixed into the print, SK Hynix and Samsung down on profit-taking, SMCI, AMD and Dell up on positioning, as global semiconductor sales hit a record $120.6 billion in May, up 104% year over year. Consensus expects Nvidia revenue to grow more than 50% year over year on Blackwell GPU shipments and hyperscaler data-centre demand.

Source: CNBC

Tuesday, 25 August 2026

Regulation

The SEC is investigating Situational Awareness, subpoenaing the banks that financed and cleared trades for Leopold Aschenbrenner's AI hedge fund after its near-collapse in July

Federal regulators have told banks that supervised the fund's trading and channelled funding to it to preserve all records, according to the New York Times. Situational Awareness has not been accused of any wrongdoing and says it will cooperate fully; the probe follows the fund being forced to unwind its public stock holdings after AI-stock declines erased billions of dollars in value at the end of July.

Source: TechCrunch

Sunday, 23 August 2026

Research

F2Agent paper proposes a multimodal fusion architecture for LLM trading agents, reporting over 20% average annualised-return improvement against 16 baseline methods across six stocks and crypto assets

Submitted 6 August 2026, the paper argues existing multimodal trading agents fail to capture cross-modal dependencies between price, text and other signal types and remain fragile to market noise. F2Agent's fix is a modality-aware fusion mechanism plus a noise-robust consistency regularizer, tested on six assets including GOOG and TSLA.

Source: arXiv

Market-structure

Robinhood extends its Agentic Trading feature to cryptocurrency, letting AI agents trade digital assets around the clock through the same dedicated, fund-limited accounts already used for equities and options

The rollout follows Robinhood crossing 100,000 agentic trading accounts in late July, on a platform that now holds roughly $350bn in total assets. Crypto trades run through the same ring-fenced Agentic Accounts structure, kept separate from a user's main brokerage portfolio and capped at whatever funds are specifically allocated to them.

Source: The Crypto Times

Institutional

Standard Signal, a Y Combinator-backed fund, launches billing itself as the first hedge fund where AI researches and executes every trade end to end, live with a reported Sharpe ratio above 3

Founded by Phind creator Michael Royzen, Standard Signal trains its own reasoning models to read order flow, market microstructure and macro data in real time and place trades within preset risk limits, with no human sign-off on individual positions. It is opening to a small number of early investors while the strategy trades live.

Source: Y Combinator

Saturday, 22 August 2026

Institutional

Ai Funds brings its BAILA stock-picking model to the public markets, launching the actively managed HIAI ETF on Cboe BZX with a 40 to 60 stock concentrated portfolio and a 0.87 percent expense ratio

Ai Funds has run its Bayesian AI Learning Algorithm for financial advisors managing hundreds of millions of dollars for several years. HIAI is the firm's first attempt to sell that same model directly to the ETF market, competing on a public, dated track record rather than a fund manager's word for it.

Source: GlobeNewswire

Institutional

Citi, HSBC, Barclays, Deutsche Bank and Standard Chartered sign on to Ant International's upgraded Falcon Time-Series Transformer, wiring one AI forecasting model into competing banks' own FX hedging platforms

Ant International says its FalconTST 2.0 model can cut foreign-exchange hedging and allocation costs by more than 60 percent by forecasting currency risk and cross-border cash flow needs. Rather than each bank building its own model, five major global banks are plugging the same third-party system into their existing FX infrastructure.

Source: Reuters

Market-structure

Binance launches Agent OS, a standardised access layer that lets ChatGPT, Claude, Codex and Cursor place spot and futures trades through permissioned sub-accounts, with withdrawals blocked by default

Binance's new developer platform bundles its API, wallet infrastructure, x402 payment rails and a Model Context Protocol server into one system so outside AI tools can read market data and trade without users managing API keys directly. The exchange tracks what an agent does but not why, and most of the risk control sits with whoever sets the permissions.

Source: TechCrunch

Friday, 21 August 2026

Regulation

The CFTC's Innovation Advisory Committee held its first meeting on 20 August, formally placing AI agents on the agenda alongside crypto and prediction markets, a day after a White House summit on the same topics with SEC and CFTC chairs

The committee's inaugural session followed through on Wednesday's regulatory summit rather than repeating it: instead of another statement of intent, the CFTC used the meeting to start scoping how autonomous AI agents interacting with derivatives and prediction markets should sit inside its existing rulebook.

Source: Bloomberg

Institutional

UBS Investment Bank and FactSet backed Finster AI's Series B round, funding an AI platform built by former DeepMind, Meta, J.P. Morgan and Morgan Stanley staff that automates research, company modelling and sector monitoring for investment banking and asset management desks

The round, announced 10 August, is a bank buying into the infrastructure layer that feeds trading and advisory decisions rather than into a trading algorithm itself, which is exactly where most of the real institutional AI money in 2026 is actually going.

Source: PR Newswire

Research

A new benchmark, KTD-Fin, anonymises tickers, dates and market narratives before testing ten frontier LLM trading agents on China's CSI300, then uses performance attribution to show most of their returns come from passive market and style exposure rather than genuine stock-picking skill

The paper's core complaint about existing LLM trading benchmarks is that long backtests overlap with a model's training-data knowledge cutoff, letting memorised prices and narratives masquerade as investment reasoning. KTD-Fin's masking protocol strips that out, and the attribution results that follow are unflattering.

Source: arXiv

Thursday, 20 August 2026

Regulation

The CFTC has sent a draft request for public comment on AI compute futures to the White House's OMB for review, formalizing oversight of a market CME and ICE are racing to build ahead of CME's planned 5 October launch

CME's Silicon Data H100 and B200 Rental Index Futures already have a target launch date; ICE is building competing contracts on separate compute-price indices. Once OMB clears the request, a 30-to-60-day public comment window opens before any contract can actually trade.

Source: crypto.news

Institutional

Hudson River Trading posted a record $11.4bn in second-quarter trading revenue and a gain for July, reportedly outperforming rival Jane Street's roughly $15bn AI-trade loss the same month

HRT's Q2 haul alone nearly matches its entire 2025 net trading revenue of $12.3bn, and first-half 2026 revenue already stands at $17.8bn combined. The contrast with Jane Street, covered on this site when the loss broke, is a case study in how differently two quant-heavy firms can come out of the same AI-stock volatility.

Source: CryptoBriefing

Market-structure

Prop trading firms including AQR, Susquehanna and Jump Trading are building infrastructure for Kalshi and Polymarket as combined prediction-market volume hit $13.7bn in June, even as a benchmark test found frontier AI models losing 16-30% when trading those markets directly

Kalshi's institutional volume climbed 800% and its annualized volume now exceeds $178bn, drawing serious quant infrastructure. But the Prediction Arena benchmark, which gave six frontier models $10,000 each to trade prediction markets from January to March, found every one of them lost money on Kalshi and roughly broke even on Polymarket.

Source: CryptoSlate

Wednesday, 19 August 2026

Research

A new arXiv paper describes AQuA, a pair of AI research agents that recursively improve their own quant strategies, reporting a 2.50 Sharpe equity strategy that stayed positive every year from 2021 through 2025

Unlike most of the trading-agent literature this site has covered, AQuA isn't an agent that trades, it's an agent that does the research work of building trading strategies, then uses its own prior results to get better at that job.

Source: arXiv (Guo, Huang, Gao, Li, Ge, Kuang, Wang)

Sunday, 16 August 2026

Institutional

Two weeks after Situational Awareness's $35bn collapse, named Silicon Valley backers including Stripe's Collison brothers, Nat Friedman and Daniel Gross are offering Leopold Aschenbrenner fresh capital, which the fund is currently declining

The fund's forced liquidation ran on roughly 400 percent leverage and triggered a Citadel fire-sale, yet it preserved private stakes in Anthropic, Fluidstack and MatX and has already deployed $500m into new private bets, a conviction-over-track-record financing pattern distinct from the collapse itself.

Source: 247wallst.com

Research

A new benchmark, Backtrader-Bench, finds tool-augmented LLM trading agents hit 90 percent accuracy on standard questions versus 73 percent without tools, but performance collapses to near-random on harder, auto-mined questions

Testing 11 models without tools plus 4 tool-augmented configurations across 5 strategies, 33 templates and 3 difficulty tiers, the paper finds tool access materially improves surface-level accuracy while doing little to fix the underlying robustness problem once question difficulty rises.

Source: arXiv (2608.11232, FinLLM Workshop @ IJCAI 2026)

Institutional

Jane Street lost an estimated $15bn in July on heavily leveraged AI trades, around 37.5 percent of its 2026 trading revenue to date, with a large share tracing to its backing of Leopold Aschenbrenner's now-liquidated Situational Awareness fund

The loss highlights a leverage chain running from private AI bets through to public AI-infrastructure stocks, with Jane Street's exposure to CoreWeave and the Nvidia financing web sitting at the center of the July drawdown.

Source: ts2.tech (aggregating Bloomberg reporting)

Saturday, 15 August 2026

Institutional

Hazeltree's securities-finance data shows hedge funds rotating within the AI trade in July, away from infrastructure exposure and toward names showing actual AI revenue, with SpaceX's long/short ratio collapsing from 2.8 to 0.55

The Data Insights Crowding Report, drawn from anonymised positioning across more than 700 hedge funds and roughly 16,000 securities, found funds made measured internal adjustments rather than broad AI de-risking in July. Tesla and Amazon gained long interest while Apple, Meta and Nvidia saw long holders decline and shorts build; SpaceX's long/short ratio fell from 2.8 in June to 0.55 in July alongside a 31 percent stock decline, and Intel flipped from net-short to net-long.

Source: Hedgeweek

Thursday, 13 August 2026

Regulation

CFTC Chairman Michael Selig published the agenda for the regulator's inaugural Innovation Advisory Committee meeting on August 20, with a dedicated session titled 'Artificial Intelligence: Preparing for Intelligent Markets' covering AI in trading, compliance, surveillance and risk management

The session will address agentic finance directly, autonomous systems that execute transactions and manage portfolios without continuous human oversight, and will ask whether the CFTC's existing regulatory principles adequately cover AI-enabled market participants. The meeting streams live from Washington, sits alongside crypto and prediction market sessions on the same agenda, and keeps a public comment window open through August 27.

Source: CFTC

Market-structure

South Korea's Kospi index entered technical bull market territory on August 13, rebounding roughly 23 percent from its July 30 low as Samsung Electronics and SK Hynix surged on renewed AI chip demand signals

The index jumped more than 4 percent at the open, with SK Hynix up over 7 percent and Samsung up more than 4 percent, driven by stronger-than-expected earnings from global tech names including Supermicro and CoreWeave that pointed to continued heavy AI infrastructure spending. The reversal comes just two weeks after leveraged AI-memory ETFs triggered the Kospi's worst monthly performance since 2008, when regulators tripled minimum deposit requirements and froze new leveraged-product listings to contain the fallout.

Source: CNBC

Market-structure

The SEC's June 16, 2026 proposal to rescind Regulation NMS Rule 611 (the Order Protection Rule) and Rule 610(e)'s locked/crossed-market prohibition would end the requirement that broker-dealers connect to every exchange and guarantee the National Best Bid and Offer on every execution, a change the industry is calling the biggest equity market-structure shift in two decades

The proposal would shift emphasis onto broker-dealers' existing best-execution obligations rather than a rule-mandated connectivity requirement, aiming to cut the cost of maintaining links to a proliferating, fragmented set of exchanges. Public comments closed August 17, 2026. The SEC's own framing weighs the benefits of execution flexibility against the risk of less uniform quoting and more decentralized execution across venues.

Source: Traders Magazine

Institutional

Whale Rock Capital, Alex Sacerdote's $19 billion Boston tech fund, saw its flagship hedge fund drop 21.7% in July as AI and semiconductor stocks reversed, cutting its 2026 gains roughly in half from 72.5% to 35.1% year-to-date

Whale Rock had ridden a chipmaker rally through the first half of 2026, adding to stakes in SanDisk, Bloom Energy and CoreWeave in Q1. All three tumbled in July as a broader memory-sector selloff hit chip and AI-infrastructure names hardest. The firm's long-only fund fell a comparatively milder 18.8%, still holding a 36.8% year-to-date gain, underscoring how much of the damage was concentrated in the hedge fund's more leveraged, concentrated positioning.

Source: Bloomberg

Research

Researchers Junyi Yao and Zihao Zheng audited 30 LLM-based trading studies across data provenance, execution timing, turnover treatment and transaction-cost modeling, finding that agent architecture is well-documented while evaluation methodology is not, and that realistic frictions materially compress reported returns

The paper scores each study on seven dimensions, point-in-time controls, split transparency, held-out evaluation, cost and turnover treatment, execution semantics, universe definition and artifact availability, building a coded evidence matrix rather than a narrative literature review. A worked 10-equity example shows that adding explicit transaction costs and realistic execution timing can substantially shrink the outperformance claimed by an active LLM strategy versus a naive backtest.

Source: arXiv

Wednesday, 12 August 2026

Institutional

Bloomberg reports Wall Street underwriters are steering AI-linked hyperscaler bond deals toward pension and insurance buyers to stop fast-money investors from tanking the debt on its first day of trading

Since the start of 2025, Alphabet, Meta, Amazon and Oracle alone have issued more than $300 billion in bonds to fund AI infrastructure, but debt from Nvidia, SpaceX and Amazon has repeatedly cratered below issue price within hours of trading as fast-trading accounts flip the paper for quick profit. Underwriters are now favouring buy-to-hold real-money accounts over those investors and spacing out issuance, with bankers privately expecting $50-60 billion more hyperscaler debt to hit the market after Labor Day.

Source: Bloomberg

Retail

Betterment's fourth annual Retail Investor Survey of 1,000 US investors found only 31 percent overall trust AI for financial advice, but 53 percent of those who do trust it say AI has already influenced a financial decision they would not otherwise have made

Gen Z investors are eight times more likely than Baby Boomers to feel comfortable using AI for long-term financial planning (41 percent versus 5 percent), and 60 percent now cite social media as their primary source of financial news, up from 45 percent in 2024. The same survey found 52 percent of Gen Z investors have redirected investing money toward sports betting in the past year, prompting Betterment CEO Sarah Levy to warn that 'when a prediction market or sportsbook starts to feel like a retirement strategy, we have a problem.'

Source: PR Newswire

Market-structure

Bloomberg analysis finds AI-related exposure inside bullish leveraged equity ETFs has climbed from 26% to 58% since 2022, with just ten stocks including Nvidia, SK Hynix and Micron accounting for two-thirds of that exposure

The data, covering average daily exposure across roughly 800 leveraged ETFs from 26 June to 6 August 2026, shows single-stock leveraged products in memory-chip makers now account for nearly half of the category's exposure on their own. Bloomberg and JPMorgan strategists flag that leveraged ETFs' share of daily trading volume far exceeds their share of assets, meaning routine rebalancing flows can move the underlying stocks more than their fundamentals do.

Source: Bloomberg

Research

Researchers Bo Qu and Mingguang Chen publish CLQT, a benchmark showing that once look-ahead leakage and transaction costs are controlled, the apparent top-performing LLM portfolio-management agent in existing tests is often just a reliability artefact from a handful of lucky rounds

The paper, submitted in June and revised 2 August 2026, reframes evaluation of LLM trading agents as diagnosis rather than ranking, running a five-stage gather-synthesise-allocate-execute-reflect cycle with verifiable audit trails and a five-axis capability scorecard. Its central finding is that the agent with the best headline Sharpe ratio differs from the agent with the best underlying capability profile, with the nominal winner in one test traced to what the authors call a five-of-26-round reliability artefact.

Source: arXiv

Retail

eToro's Q2 2026 results show net income up 77% to $53 million alongside a $231 million deal to acquire US broker TradeZero, while its new Agent Portfolios product with Claude and Grok connectors is so far being adopted mainly by sophisticated users, not the mass retail base

Net contribution rose 9% year over year to $229 million and adjusted EBITDA rose 9% to $78 million, with the flagship Alpha Portfolio, built on eToro's proprietary trading data, returning more than 30% in the first half of 2026 and Smart Portfolio assets growing 58% year over year. Agent Portfolios lets customers connect AI agents to designated portfolios with their own objectives, capital allocation and risk parameters, sitting alongside eToro's AI app store, which now carries more than 75 third-party applications.

Source: eToro / GlobeNewswire

Research

Researchers at Samaya AI released FrontierFinance, a fully open benchmark of 220 expert-crafted queries and 11,543 source-attributed rubrics spanning the full investment analyst workflow, built specifically because existing finance benchmarks have been largely saturated by current models

Where prior benchmarks mainly tested narrow financial data extraction, a task current frontier models have already mastered, FrontierFinance spans six use cases including financial modeling, sector and macro analysis, earnings and events, company research, catalyst monitoring, and screening, scored against thousands of source-attributed rubric items rather than generic LLM-as-judge grading. The authors position it as both broader and harder than existing public finance benchmarks, aimed at open-ended, long-form analyst queries that reference-based metrics struggle to score.

Source: arXiv

Regulation

Reuters reports Kalshi has signed a multi-year deal to run Nasdaq's algorithmic market-surveillance platform across its prediction markets and perpetual-style futures, after regulators fined former Congressman George Santos and opened an investigation into a White House staffer over alleged insider trading on the exchange

Nasdaq's surveillance technology, already used by more than 50 exchanges and 20 regulators globally including the CFTC itself since August 2025, will give Kalshi real-time cross-market pattern detection and format trade data for direct CFTC submission, layered on top of the Solidus Labs surveillance system Kalshi added in February. The deal lands as Kalshi's perpetual futures volume hit $16.1 billion cumulative by early July and prediction markets face mounting congressional scrutiny over insider trading enforcement.

Source: Reuters

Tuesday, 11 August 2026

Institutional

Broadridge's fiscal 2026 earnings call detailed an agentic AI partnership model offering managed-services clients up to 30 percent day-one operational cost reduction, alongside a projected $25 million in internal AI-driven productivity savings for fiscal 2027

CEO Timothy Gokey positioned Broadridge as 'the agentic operating system' for capital markets clients, with AI agents that analyze real-time operational data, identify exceptions and initiate resolution autonomously. Platform-enabled sales, including AI-powered products, rose 60 percent year over year and now represent nearly 40 percent of total closed sales, while the firm's distributed ledger repo volume tripled to $360 billion daily since May 2025.

Source: The Motley Fool

Market-structure

Indonesian multi-asset platform Pluang launched agentic trading in controlled beta, letting whitelisted users connect ChatGPT, Claude or Gemini directly to live regulated accounts to research, propose and execute trades across stocks, gold, money market instruments and crypto, with public rollout set for mid-August

Every AI-proposed transaction still requires explicit user confirmation before execution, and cofounder Claudia Kolonas said safeguards "must be embedded at infrastructure level, not depend only on model algorithms." The move follows similar 2026 launches from Robinhood, Interactive Brokers and TradeStation in the US, and puts Indonesia among the first Southeast Asian markets to open a regulated brokerage's execution layer to third-party general-purpose AI assistants.

Source: The Jakarta Post

Retail

Charles Schwab data shared with Axios shows retail investors stayed net buyers through July's AI selloff, with Gen X clients and self-directed traders driving the heaviest buying in beaten-down names like Micron and SanDisk even as a separate Schwab sentiment survey found clients notably less bullish on stocks overall

Schwab's Trading Activity Index rose to 59.80 in July from 59.12 in June, and SpaceX drew the most consistent buying interest of any name in the month. Technology was a net sell sector overall even as investors bought the dip in specific AI-linked chip names, a split Schwab's head trading strategist summarised as clients wanting AI exposure "on sale."

Source: Axios

Institutional

Value Aligned Research Advisors' $20 billion AI-only hedge fund lost 44% in July, with Bloomberg reporting the Princeton firm's flagship VAR AI Fund was hit by the same AI infrastructure selloff that forced rival Situational Awareness into a distressed sale, even as VARA's year-to-date return held at 65%

VARA, run by Ben Hoskin and David Field out of Princeton, New Jersey, holds roughly $15 billion in the hedge fund strategy and $5 billion in a long-only account, all concentrated in AI-centric equities with no crypto exposure. The fund was up 90% through April and 113% for all of 2025 before July's reversal, and Bloomberg links part of the turbulence directly to the market disruption caused by Situational Awareness's forced unwind.

Source: Bloomberg

Monday, 10 August 2026

Research

A multi-agent LLM system called Agora reached a long-short Sharpe ratio of +1.87 on a sealed 91-day CSI 1000 holdout, beating its closest baseline's best-seed result of +1.33, but the authors flag their own headline number as a single, unrepeated run

Five specialised LLM roles jointly discover and refine trading factors across eight evolving skill libraries with information barriers to the holdout set. The baseline's seed variance tells the more important story: mean performance across three seeds dropped to -0.755, and Agora's own full-system seed stability remains uncharacterised.

Source: arXiv (Panda AI)

Retail

Mirae Asset Securities' AI-driven pension robo-advisor cut personal pension losses to around -10% for subscribers during July's 22% KOSPI plunge, against roughly -61% for a comparable non-user cited in the same report

Subscription accounts nearly tripled year over year to 141,381, with assets under management up 46% to 8.34 trillion won. The service expanded bond weighting to 8.2% during the volatility, a defensive shift credited with much of the outperformance.

Source: Seoul Economic Daily

Regulation

Securities class action filings tied to AI arrived at nearly twice the 2025 rate through the first half of 2026, with NERA counting 18 new AI cases and Cornerstone tallying 15, split across AI development, data centres and infrastructure claims

Aggregate settlements topped $2 billion by midyear, already 75% of the inflation-adjusted total for all of 2025, driven by outsized cases like Didi Global's $740 million IPO investor settlement. Roughly 120 new proposed securities class actions were filed overall in H1, putting 2026 on pace to exceed last year.

Source: Bloomberg Law

Sunday, 9 August 2026

Retail

More than 5,000 retail investors are still using the copy-trading platform Autopilot to mirror Leopold Aschenbrenner's publicly disclosed stock picks, and their unlevered version of his portfolio is up 53.4% since March while his actual hedge fund lost 67% in July

Autopilot replicates Situational Awareness's disclosed AI-infrastructure holdings, Nebius, Coreweave and Bloom Energy among them, without leverage, shorts or derivatives. CEO Brian Schardt's summary of why his copiers didn't blow up alongside the fund they're copying: 'Ours is not leveraged, which means we don't take the same risk that he took, which is better.'

Source: AOL (Business Insider)

Market-structure

BestEx Research launched Pulse AI, a conversational interface that plugs into Claude, ChatGPT and other AI assistants so institutional traders can query pre-trade liquidity analysis and, from Q1 2027, broker-neutral transaction cost analysis in plain language

Pre-trade analytics for global futures are live now, with equities following in October. CEO Hitesh Mittal's framing of the problem this solves: TCA 'has remained rigid and complex for decades,' locked into quarterly reports that arrive too late to change how a desk trades in the moment.

Source: PR Newswire

Institutional

Jamie Dimon told CNBC that margin debt is the highest it has ever been, days after reporting confirmed JPMorgan was one of the prime brokers behind Leopold Aschenbrenner's AI hedge fund Situational Awareness before its leveraged collapse

Dimon named four sources of leverage regulators can't fully see, prime brokerages, hedge funds, leveraged ETFs and Treasury arbitrage trades, warning that 'somebody will disrupt the market in a quick way' before anyone spots it coming. JPMorgan's own brokerage relationship with the fund that just did exactly that went unmentioned in his warning.

Source: Yahoo Finance

Research

An NBER survey of nearly 6,000 executives across the US, UK, Germany and Australia finds 69% of firms actively use AI, yet nine-in-ten report no measurable impact on productivity or employment over the past three years

Executives personally using AI average just 1.5 hours a week with it. Looking ahead, they expect AI to lift labour productivity by only 1.4% and cut employment by 0.7% over the next three years, a modest number now being cited on both sides of the argument over whether this year's AI capex is justified.

Source: National Bureau of Economic Research

Regulation

The SEC's Division of Examinations has begun requesting AI-related materials from financial firms as part of a sweep targeting AI-driven portfolio management, algorithmic trading models and marketing claims that firms may not be able to substantiate

The requests span written, audio and video materials mentioning AI, ADV Part 2 filings, websites and pitch decks. The Division is specifically probing what compliance professionals call 'AI washing,' public claims about AI capability that the underlying systems don't actually support.

Source: FinTech Global

Saturday, 8 August 2026

Institutional

A coordinated AI voice-cloning campaign hit Citadel, Point72, Two Sigma and Millennium Management within days of each other on August 5, forcing the first live test of FINRA's five-month-old Financial Intelligence Fusion Center

Attackers used AI to clone the voice, tone and phrasing of trusted executives in phone calls and audio messages, trying to talk employees into handing over credentials or network access. Two Sigma says it blocked the attempt outright, Point72 found no evidence of stolen client data so far, and Citadel and Millennium are not saying anything at all.

Source: Fortune

Research

A 20-year backtest across more than 100 stocks by researchers from Edinburgh, UCLA and Sungkyunkwan found buy-and-hold beat every one of the 20-plus AI trading strategies tested on a risk-adjusted basis, ahead of the team's presentation at ACM SIGKDD this week

On steadier, low-volatility stocks, plain buy-and-hold returned about 7.9% a year and topped the field once risk was accounted for. One AI strategy posted a raw 14% return but with swings large enough to lose on a risk-adjusted basis, and a 55-year-old ARIMA statistical model outperformed most of the AI tools tested.

Source: Yahoo Finance

Regulation

The EU's three financial regulators, EBA, EIOPA and ESMA, issued a joint statement warning that frontier AI models are compressing the gap between vulnerability discovery and exploit to the point that existing DORA-era cyber controls may not keep pace

The Joint Committee's 31 July statement does not create new rules. It reframes cyber resilience under the Digital Operational Resilience Act as an AI problem, telling banks, investment firms, fund managers and insurers to update governance around a threat that moves faster than their existing incident-response cycles.

Source: European Banking Authority

Market-structure

MetaMask opened general access to Agent Wallet on August 6, a self-custodial wallet built to let Claude Code, Codex, Cursor and other coding agents execute on-chain trades through a CLI, constrained by spending limits the user sets in advance

Agent Wallet gives an AI coding agent its own wallet rather than the user's private keys, then enforces those limits through simulation, threat scanning and a mode called Guard Mode that pauses anything outside policy for two-factor human approval. An opt-in Beast Mode trims those interruptions for users willing to trade oversight for speed.

Source: MetaMask

Friday, 7 August 2026

Regulation

California's AB 316, in force since 1 January 2026, bars any business that developed, modified or deployed an AI system from arguing the AI "acted autonomously" to escape liability, a defence collapse that lands squarely on firms operating money-moving trading agents

The law does not create new strict liability for AI harm, it removes exactly one argument, that autonomous behaviour breaks the chain of responsibility, while leaving every other defence intact. For anyone deploying an agent that places trades, the practical read is that every transaction is now attributed to a person, and the plaintiff gets to pick which one.

Source: Astraea Counsel

Institutional

Quantedge's $7.2bn cross-asset systematic fund gained 7.8% in July and sits up 34.6% for the year, turning the same AI selloff that cost Millennium 2.1% and Jupiter-Minghong more than 40% into its strongest month of 2026

Every hedge fund headline this week has been a variation on losses, Millennium, Coatue, Jupiter-Minghong, all hurt by concentrated AI exposure. Quantedge's gain is the other half of the same story, a reminder that the July selloff was a test of structure and correlation, not a uniform tax on anything AI-adjacent.

Source: Hedgeweek

Market-structure

The Philadelphia Semiconductor Index fell 21% in July, its worst month since October 2008, yet the final week alone pulled in $12 billion of net ETF inflows, a quarter of all US ETF money that week into funds holding 1% of total ETF assets

Nearly half of July's trading sessions closed with 4%-plus moves and every session saw at least a 2% intraday swing, the kind of volatility profile last seen in 2020. Investors did not treat that as a reason to leave, they treated it as the buying opportunity of the year.

Source: Fortune

Research

A PRISMA-compliant review screening 847 records down to 60 qualifying studies finds the academic literature still cannot agree on whether AI-driven algorithmic trading improves or degrades market liquidity, price discovery and stability, only that it changes all three

Fuli Yang, Shiyu Lin and Jing Huang's systematic literature review, published on SSRN 15 July, is less a verdict than a map of how unsettled the underlying research still is, which matters more than another single study's headline finding for anyone trying to reason about what widespread AI trading actually does to market structure.

Source: SSRN

Thursday, 6 August 2026

Institutional

JPMorgan's industry-wide numbers put July's AI unwind at roughly 3 percentage points off hedge funds as a whole, with multi-strategy losing 2.2% while Asia-Pacific equity long-short lost 9.4%, its steepest decline of any strategy tracked

The single-fund stories this week, Millennium's 2.1% and Jupiter-Minghong's 40%+, both sit inside this same JPMorgan dataset, and the spread between them is the real story: the same July AI selloff produced wildly different outcomes depending on strategy structure and leverage, not just exposure to AI names.

Source: Hedgeweek (JPMorgan analysis)

Wednesday, 5 August 2026

Retail

Bloomberg's profile of retail traders building AI-assisted automated trading systems lands on an unglamorous conclusion, that a year spent replicating hedge-fund-style options strategies with off-the-shelf AI tools mostly produced results not much better than parking the same money in an index fund

The piece follows individual traders using AI coding and quantitative tools once reserved for institutions to build automated strategies, and finds that access to the tools narrowed the gap in mechanics without narrowing the gap in outcomes, with amplified mistakes and risk-management gaps offsetting the sophistication gains.

Source: Bloomberg

Institutional

A Hong Kong quant fund with close ties to Shanghai's Minghong Investment lost more than 40% of its value in July after running two-to-four-times leverage on the same AI and small-cap momentum names that carried the fund to a 60% gain in the first five months of the year

The fund's collapse traces the same mechanism as the broader China quant carnage this summer, leverage that amplifies momentum in both directions turning a crowded AI trade from the year's best-performing strategy into one of its worst inside a single month.

Source: Bloomberg

Market-structure

Nasdaq's Global Trading Hours initiative, approved by the SEC in April, is now targeting Sunday December 6 as the launch date for 23-hour trading five days a week, with a new 9pm-to-4am session added on top of the exchange's existing pre- and post-market windows

The design keeps the 9:30am-to-4pm session as the sole price-discovery window with the Opening and Closing Crosses, meaning the new overnight session runs on continuous trading rather than auctions, an important detail for anyone building AI-driven execution logic around it.

Source: Alston & Bird

Regulation

The SEC's amended Rule 605 execution-quality disclosure requirements took effect August 1, pulling broker-dealers running alternative trading systems and single-dealer platforms into the same monthly reporting regime that previously applied only to market centers

The amendments widen what counts as a reportable venue, add fractional-share and odd-lot order categories that AI-driven retail flow increasingly produces, and introduce a new monthly summary report, with the first disclosures due under the new rules by the end of September.

Source: Federal Register / SEC

Research

A new SSRN working paper builds the first structured taxonomy of LLM and multi-agent trading systems and uses it to name four open research gaps, the most consequential being that nobody has yet built a rigorous, cost-adjusted benchmark for comparing LLM-generated trading signals against each other

The taxonomy sorts existing work along five dimensions, the LLM's functional role, information modality, learning paradigm, single-agent versus coordinated multi-agent architecture, and evaluation standard, and finds that most published results sit at the weakest end of that last dimension, in-sample backtests rather than live-market evaluation.

Source: SSRN

Tuesday, 4 August 2026

Regulation

FINRA's 2026 Annual Regulatory Oversight Report adds a standalone GenAI section for the first time, telling member firms that generic compliance controls no longer cover what their AI tools are doing inside trading algorithms and customer communications

The report names summarization and information extraction as the most common GenAI use case firms report, then walks through what supervision actually has to look like: testing for hallucinations and bias, ongoing output monitoring, vendor diligence, and books-and-records capture of AI-generated communications, all tailored to GenAI rather than bolted onto existing programs.

Source: FINRA / McGuireWoods client alert

Research

A new UCL framework for modeling AI trading agents finds that systemic risk does not rise smoothly as autonomy increases, it rises non-linearly once autonomy and infrastructure concentration outpace how well supervisors can actually observe what the agents are doing

Hui Gong's Agentic Financial Market Model links five design parameters, autonomy depth, model heterogeneity, execution coupling, infrastructure concentration, and supervisory observability, to market-level outcomes, and concludes the least risky near-term equilibrium is "bounded autonomy": AI agents as supervised co-pilots rather than fully independent market actors.

Source: arXiv (UCL Institute of Finance & Technology)

Institutional

Millennium Management and Coatue Management both posted their worst monthly results in over a year in July as the AI stock selloff whipsawed even the multi-strategy giants that were supposed to be diversified against exactly this kind of concentrated drawdown

Millennium's headline loss and Coatue's steeper drop both trace to the same source, crowded exposure to the same handful of AI infrastructure and semiconductor names across supposedly uncorrelated pods and strategies, a reminder that diversification only protects against risks that are actually distributed.

Source: Bloomberg

Market-structure

Nasdaq is rolling AI capabilities out across its market surveillance platform to every customer, not just its own exchange, after a pilot with Saudi Arabia's Capital Markets Authority caught 80% of pump-and-dump schemes that older rule-based detection missed

The platform already underpins surveillance for roughly 50 exchanges and 20 regulators worldwide, meaning an AI upgrade here changes market-abuse detection standards well beyond Nasdaq's own order book, at the same moment bad actors get access to the same generative tools to make manipulation schemes harder to pattern-match in the first place.

Source: Nasdaq, Inc.

Retail

Podium Markets AI built Ivy to read a retail investor's entire portfolio across every brokerage they hold an account at, then recommend trades, but deliberately stopped short of letting it place them, on the explicit bet that investors want the analysis without giving up the decision

CEO Dirk Mueller-Ingrand frames it as "the AI informs, but the human decides," a narrower and more cautious design than the fully autonomous execution agents Robinhood, Public, Coinbase and eToro have all shipped this year, betting that trust in cross-account AI recommendations has to be earned before it is automated away.

Source: CNBC / Podium Markets AI

Monday, 3 August 2026

Market-structure

CME Group is launching the first standardized compute futures contract later this year, letting traders hedge GPU rental prices the same way they already hedge oil, meaning AI infrastructure itself is about to become a directly tradable commodity

Built with Silicon Data, the DRW-backed GPU market data provider, the contracts are priced off daily benchmark indices for on-demand GPU rental rates, giving AI developers, cloud providers, and financial institutions a way to manage exposure to compute costs the way an airline hedges fuel.

Source: Yahoo Finance / CME Group

Regulation

The EU AI Act's Article 50 transparency rules became enforceable on 2 August, and unlike the high-risk provisions that got pushed to 2027, the disclosure duties covering AI chatbots and generated content did not move

Firms offering EU clients an AI trading assistant, portfolio chatbot, or AI-generated market commentary now have a live legal obligation to disclose that a person is interacting with AI, separate from and unaffected by the broader high-risk AI Act deadlines that the Commission's Omnibus package deferred.

Source: Holland & Knight / Gibson Dunn legal analysis

Institutional

Goldman Sachs surveyed 341 hedge fund allocators managing $1.5 trillion and found the industry's 7% first-half return, nearly double the 10-year average, came from correctly rotating through three different phases of the AI trade rather than just holding it

Almost half of allocators plan to increase hedge fund exposure in the second half of 2026 and only 3% plan to cut it, with equity long-short managers posting 12.9% gains as funds moved money from semiconductors to power and data centers and then into memory chips as each phase of the AI trade played out.

Source: Reuters (via Goldman Sachs client note)

Research

Researchers put autonomous LLM agents through classic experimental asset-market auctions and found they reproduce the same disposition effect and recency bias human traders show, and that specific prompt wording can be tuned to deliberately amplify or suppress how large the resulting bubble gets

Shumiao Ouyang and Pengfei Sui's study scored twenty distinct behavioral mechanisms in LLM trading agents' reasoning and found the same excess-demand-predicts-price and disagreement-predicts-volume relationships that decades of human experimental finance has documented, with the added finding that these tendencies are not fixed but causally steerable through prompting.

Source: arXiv (Ouyang & Sui)

Retail

Questrade became the first Canadian brokerage to let customers connect their live account to Claude or ChatGPT directly via Model Context Protocol, letting an external AI assistant read real portfolio data and draft trades for in-app approval

CEO Edward Kholodenko unveiled the MCP integration alongside commission-free US options and pre-IPO access at a Toronto event, positioning it against Wealthsimple's prediction-market push and Robinhood's in-house agentic trading account as brokerages compete on how deeply AI gets to touch a real account.

Source: The Globe and Mail

Sunday, 2 August 2026

Research

Bloomberg's head of market structure research says trading desks are closer to real agentic AI than at any point in four decades of false starts, but the honest read is that the first wave is workflow augmentation, not autonomous execution

Larry Tabb told Traders Magazine that despite repeated attempts at desk automation since the late 1980s, current agentic AI is the closest the industry has come, while cautioning that "the promise of Agentic is large, but the reality is uncertain" and that infrastructure has to be built before automation can expand.

Source: Traders Magazine

Regulation

A FINRA-style regulator for frontier AI has gone from a single Lawfare essay to a Treasury-developed proposal under White House review in four months, with Google, Anthropic and OpenAI already positioning for it

Mark Thomas's April proposal for a federally-supervised self-regulatory organisation modelled on FINRA has picked up backing from Google's own "FARO" framework, a public essay from DeepMind CEO Demis Hassabis, and now a Treasury Secretary Scott Bessent plan reportedly under active White House review.

Source: Lawfare

Institutional

Goldman Sachs Asset Management is betting AlphaAI can solve what its own $40.5 billion AI-themed ETF category cannot, which is telling investors which companies are actually converting AI into margin rather than just riding the theme

Goldman shares rose roughly 3.2% after the asset management arm launched AlphaAI, an internal platform led by Lou D'Ambrosio that folds AI into investment research, portfolio management and value-creation work, though Goldman disclosed no specific financial targets or committed assets.

Source: TradingView / GuruFocus

Market-structure

Tradeweb's CEO says the operational risk on Wall Street trading venues has shifted from a rogue trader's fat-finger to a rogue AI agent executing at machine speed, and the safeguards built for the old failure mode were not built for the new one

Billy Hult, Tradeweb's CEO since 2023 and with the venue since its founding, describes the emerging risk as not a single bad trade from a human error but an AI agent generating milliseconds of erroneous activity before anyone can intervene.

Source: Yahoo Finance

Retail

Wealth managers now compete with their own clients' AI chatbots for investment advice, and CNBC's reporting suggests the mass-affluent segment most exposed to that competition is exactly where advisory fees are hardest to justify

Wealth management leaders told CNBC's Inside Wealth that even high-net-worth clients are bringing portfolio and tax questions to AI chatbots first, a practice advisors say sharpens some client conversations but also introduces new risk around hallucinated advice and leaked personal financial information.

Source: CNBC

Saturday, 1 August 2026

Research

ECB simulations found that reinforcement-learning trading agents coordinate like a bank run under stress, while LLM-based agents behave less predictably, meaning the architecture choice itself is now a financial stability variable

Current financial supervision is built to watch institutions and products, not the internal design of the AI systems trading inside them, but the ECB's research suggests two AI agents given the same market conditions can produce meaningfully different systemic outcomes depending on how they are built.

Source: European Central Bank

Market-structure

A single thinly-traded share in a Korean pre-market session moved a tokenized SK Hynix perpetual 19% on Hyperliquid and forced $60 million of positions closed in two minutes, exposing what always-on crypto-native equity trading actually depends on

Trade.xyz's oracle worked exactly as designed, relaying a real but outsized off-exchange print into a mark price with no depth behind it, and is now discretionarily covering the resulting losses while it rebuilds pricing around its own order books.

Source: CoinDesk

Friday, 31 July 2026

Research

A live-market benchmark of LLM trading agents across GPT-4o, GPT-4.1, Claude Sonnet 4 and Gemini 2.0 Flash found that agent architecture, not the underlying model, is what determines whether a trading agent behaves aggressively or conservatively

Agent Market Arena, presented at the ACM Web Conference 2026, ran four distinct agent frameworks across five LLM backbones in live crypto and stock markets and found behavioural style tracked the framework's design, not which model was powering it.

Source: ACM Web Conference 2026 (WWW '26)

Retail

Robinhood's Agentic Trading has passed 100,000 funded accounts since its May launch and now computes live technical indicators (RSI, MACD, Bollinger Bands and more) on request, but CEO Vlad Tenev admits the underlying models sometimes push back when asked to execute

Robinhood added live indicator computation to its AI trading agent, moving it from a chat layer over static charts to something closer to a real analyst, while its own CEO is candid that the model's reluctance to trade on command is an open problem, not a solved one.

Source: Benzinga

Institutional

Leopold Aschenbrenner's Situational Awareness fund sold its entire ~$20bn public equities book to Citadel after margin calls from Goldman Sachs, JPMorgan and Bank of America wiped out the leverage behind a 439% first-half return

A fund that returned 439% in H1 2026 on concentrated, leveraged AI-stock bets lost 35-47% of those same positions in July's rout, forcing a sale to Citadel rather than a managed exit, the clearest live case study yet of AI-trade crowding risk.

Source: Forbes

Thursday, 30 July 2026

Market-structure

The Bank of England's Prudential Regulation Authority has opened a review of London-based prime brokers' exposure to a narrow set of AI-linked Asian equities, after hedge funds' global prime brokerage balances rose roughly 40% over the past year

The review follows the BoE's July 2026 Financial Stability Report warning that AI-driven gains have concentrated in a handful of companies, and comes as Goldman Sachs discloses that 16% of its prime brokerage business is directly exposed to AI memory-related stocks like Samsung and SK Hynix.

Source: Bloomberg

Institutional

Goldman Sachs, JPMorgan and other Wall Street prime brokers are demanding more collateral from hedge funds whose books are concentrated in AI-related names, after the Nasdaq 100 fell 10% from its June peak

SanDisk and Intel are down 53% and 39% respectively since the peak, the Philadelphia Semiconductor Index is 25% lower, and Goldman disclosed that AI storage-chip exposure now makes up 16% of its prime brokerage book, with many of the collateral calls triggered automatically by contractual risk provisions rather than discretionary bank decisions.

Source: Financial Times (via TradingView)

Regulation

House Financial Services Committee Democrats, led by Reps Bill Foster and Brad Sherman, sent SEC Chair Paul Atkins a letter demanding answers on AI trading agent oversight, with a response deadline of 31 July

The letter argues that AI firms building trading agents have "operated largely outside the securities regulatory framework, even though their systems are making or enabling consequential investment decisions on behalf of retail investors," and asks whether the SEC already has the legal authority to police them or needs Congress to act first.

Source: Yahoo Finance

Research

A new NBER working paper by Ralph Koijen and Bradford Levy finds that optimised agentic AI systems extracting signals from earnings call transcripts more than double the explained variation in stock returns around earnings announcements, pushing R-squared from roughly 8% to approximately 20%

The paper builds a real-time, out-of-sample benchmark that evaluates AI systems using only information available at the time, avoiding look-ahead bias, and the authors argue the approach's real value is producing "human-readable economic mechanisms" that can be compared against existing asset pricing models to identify previously unexplained sources of return variation.

Source: National Bureau of Economic Research

Retail

Robinhood, Public and startup Podium Markets AI are building AI agents that manage retail portfolios and trade around the clock, with Citizens fintech research head Devin Ryan predicting "effectively everybody has their own family office that is working 24/7 for them"

Robinhood has opened its platform to third-party AI agents connecting directly to customer accounts, Public is building proprietary agents inside its own ecosystem, and Podium Markets AI's assistant Ivy analyses a customer's holdings across multiple brokerages to generate recommendations, with Ryan estimating an investor placing two trades a month today could see an AI system execute 20 in a single day.

Source: CNBC

Wednesday, 29 July 2026

Research

A new theoretical model combining performative prediction, algorithmic herding and cognitive dependency argues that widespread AI adoption paradoxically thins market depth by making supposedly independent trading signals converge, withdrawing liquidity exactly when stress makes it most needed

Researchers Shuchen Meng and Xupeng Chen build a unified framework showing that AI systems trained on similar data and comparable methods develop overlapping common-signal interpretations that collapse diverse strategies into herd-like positioning, while AI predictions reshape the very fundamentals they forecast in a self-reinforcing loop that decouples price discovery from underlying value.

Source: arXiv

Regulation

FCA chief executive Nikhil Rathi told techUK's AI-in-financial-services conference the regulator will not write a comprehensive AI rulebook because legislation cannot keep pace with technology cycling every three to six months, opting for judgement-led supervision instead

Speaking on 24 June 2026 at techUK's Agents of Change event, with more than 80% of UK financial firms already using AI in some form, Rathi set out an approach built on selective rule-writing, an agentic-AI surveillance layer for wholesale markets, and more frequent use of the FCA's existing competition powers rather than a new detailed code.

Source: FCA

Institutional

Goldman Sachs surveyed 341 hedge fund allocators overseeing $1.5 trillion and found record net demand for the strategy heading into the second half of 2026, with nearly half planning to add exposure and only 3% planning to cut it

The July poll showed institutional investors booked 7.3% average first-half returns and private capital investors 8.8%, with every major hedge fund strategy pulling in fresh money simultaneously for the first time in five years and multi-strategy funds recording their strongest inflows over the same period, momentum the bank attributes to the AI stock rally.

Source: The Standard (HK)

Market-structure

Global semiconductor stocks have shed $3.3 trillion since June as AMD fell roughly 5% and Intel nearly 6% in mid-July, while the Russell 2000 and defensive sectors rallied, the clearest sign yet that this year's AI-driven bull market is rotating rather than simply cooling

TSMC's raised 2026 capital-expenditure guidance of $60-64 billion was read as a margin-compression warning rather than good news, GPU rental prices fell roughly 31% over three weeks, and capital moved into small-caps, industrials, energy, financials and real estate even as the Nasdaq slipped 2.9% for the week ending 17 July.

Source: ClaritX

Retail

TradeZing launched TradeZ, a conversational AI trading copilot that scores the historical accuracy of the analysts and influencers it summarises before routing users to one-tap execution across more than 15 brokerages

Rolling out in phases to TradeZing's $9.99-a-month PRO members from 21 April 2026, TradeZ's headline feature is Insight Accuracy, a proprietary score tracking how often a given analyst, influencer or outlet has been directionally right over time, layered onto natural-language trade alerts with entry and exit levels.

Source: Yahoo Finance

Tuesday, 28 July 2026

Retail

eToro rebuilds its flagship app around AI agents that trade continuously inside their own sub-accounts, a structural shift from AI-as-feature to AI-as-account-holder

Launched 7 July at a London event for eToro's 40 million users across 75 countries, the new app lets investors build or copy AI trading agents that run around the clock inside dedicated sub-accounts, while a second AI agent called Tori surfaces portfolio insights unprompted; CEO Yoni Assia frames the underlying model as built on the real decisions and track records of millions of investors rather than a generic system.

Source: GlobeNewswire / eToro

Institutional

Industry survey data puts hard numbers on what AI adoption actually means inside hedge funds: 74% for operational grunt work, 69% as a co-pilot, just 6% making the final trade decision

Hedge fund assets under management reached $5.6 trillion in Q2 2026 after a record $409 billion quarterly increase, and industry voices now describe AI adoption as a baseline rather than a differentiator, but the split between where funds let models operate with real autonomy versus where a human still signs off is the more useful number for gauging how much of that adoption is substantive.

Source: Hedgeweek

Regulation

HM Treasury's financial services AI adoption plan puts the FCA on the hook for a review of LLM-generated financial guidance, not just agentic trading systems

The ten-point roadmap, drawn up by AI Champions from Starling Bank and Lloyds Banking Group and accepted by HM Treasury on 14 July, spans five areas from regulatory clarity to agentic payments, but its most consequential line item may be the instruction for the FCA to formally assess what happens when consumers take financial guidance from general-purpose chatbots that were never authorised to give it.

Source: GOV.UK / HM Treasury

Research

A backtested hybrid trading system combining technical indicators, FinBERT sentiment and XGBoost claims a 135% two-year return beating the S&P 500, the kind of result that is easy to publish and hard to trust

The preprint, accepted for the ComSIA 2026 conference, layers trend and mean-reversion signals with machine-learning trade calls and a volatility-based regime filter, reporting a final portfolio value of $235,492.83 over 24 months against a starting stake; the number is real, the question is whether the backtest methodology behind it would survive contact with live capital and an untested market regime.

Source: arXiv preprint

Market-structure

NVIDIA reportedly guaranteeing up to $250bn in financing for OpenAI's data-centre buildout knocked AMD down 8% and Intel down 4% in a single session, the market pricing vendor financing as a demand-quality problem

The Wall Street Journal reported Monday that NVIDIA is in talks to backstop OpenAI's compute expansion at a $250 billion scale, and the selloff across chipmakers that followed was not really about NVIDIA's own numbers, which stayed strong at $75.25bn in Q1 FY2027 data-centre revenue, up 92% year-on-year; it was about whether a seller financing its own biggest customer's purchases can still count as organic demand.

Source: Yahoo Finance

Monday, 27 July 2026

Market-structure

One hyperscaler's capex guidance dragged the Nasdaq down 2.2% on 24 July, a live case study in the index-concentration risk regulators keep warning about

Alphabet beat earnings but raised its 2026 capex forecast to $195bn-$205bn from $180bn-$190bn, and its shares fell over 7% on the news, its worst day in more than a year, pulling Communication Services down 5.2% and Consumer Discretionary down 5.1% while only Industrials gained, a sector-level move triggered by one company's spending disclosure.

Source: Yahoo Finance

Institutional

South Korea's National Pension Service turned net buyer of Kospi stocks in July, its first month of net buying after six straight months of selling into the AI rally

The NPS and other Korean pension funds bought a net 68.4bn won ($46.8m) of domestic equities through 24 July, concentrated in SK hynix and S-Oil while trimming SK Square and Samsung names, after the Kospi's AI-fuelled run from roughly 4,300 points to above 9,000 by mid-June pushed the fund's target domestic equity allocation up twice this year.

Source: Korea JoongAng Daily

Research

A new NBER benchmark finds optimized agentic AI systems more than double the variation in stock returns they can explain around earnings announcements

Working Paper 35431 by Ralph Koijen and Bradford Levy proposes a real-time, out-of-sample benchmark that sidesteps look-ahead bias and market reflexivity, finding the best-optimized agentic AI systems push explained return variation from an R-squared of roughly 8% to close to 20% versus standard benchmarks, while also producing human-readable economic explanations for the moves.

Source: NBER

Regulation

India's central bank closed public consultation on its first AI-specific model risk framework on 24 July, replacing rules written in 2002

The Reserve Bank of India's draft Guidance on Regulatory Principles for Model Risk Management, released 24 June 2026, sets governance expectations for AI and machine-learning models across nearly the entire RBI-regulated financial sector, banks, NBFCs and beyond, and would retire a credit-risk-model framework that predates modern algorithmic trading entirely.

Source: Vinod Kothari Consultants

Retail

Charles Schwab plans an AI assistant across equities, options, fixed income, mutual funds and ETFs for H2 2026, but starts with answers, not trades

Schwab's conversational AI assistant is scheduled for the second half of 2026 and will initially focus on market questions and portfolio summaries rather than executing trades, a deliberately more conservative rollout than Robinhood's fully autonomous agentic accounts or Coinbase's AI-agent trading, both already live since June.

Source: Yahoo Finance

Sunday, 26 July 2026

Saturday, 25 July 2026

Research

A systematic review of 77 studies on LLM trading agents found only 2 reported a proper experimental protocol, meaning most of the field currently cannot be independently checked

Researchers auditing the LLM-agent-trading literature found 15 of 19 primary empirical studies sat at the lowest reproducibility tier, with almost none disclosing transaction-cost assumptions or addressing survivorship bias, a gap that matters now that agentic tools have moved from papers into live brokerage accounts.

Source: arXiv

Friday, 24 July 2026

Market-structure

A $67 billion fund's strategist has found that when retail investors sell AI chip stocks, that has been the buy signal, every time, since February

Citadel strategist Scott Rubner flagged that retail investors turned net sellers of semiconductor stocks on two July down days as the Philadelphia Semiconductor Index fell around 5% off its June all-time high, noting that every retail-selling episode he has tracked since February has been followed by a chip rally over the next five to ten days, with a median gain near 18%.

Source: Yahoo Finance

Research

Ten leading AI models were given identical starting capital and set loose on live foreign exchange markets, and the results split further apart than expected

HKU Business School's Artificial Intelligence Evaluation Lab connected 10 leading LLMs, including models from OpenAI, Anthropic, Google, Alibaba, Moonshot AI, Zhipu AI, ByteDance and DeepSeek, to live FX data through its Agentic Trader system, with identical starting capital in an identical market environment, and found Alibaba's Qwen produced the strongest profit.

Source: HKU Business School

Institutional

An 18-year-old listed fund just posted its best year ever, and artificial intelligence exposure is one of only four themes it credits

WAM Active, the Wilson Asset Management-run listed investment company, reported a 75.5% portfolio return for the twelve months to 30 June 2026, its strongest result in an 18-year history, crediting exposure to critical minerals, electrification and grid infrastructure, precious metals and artificial intelligence as the four themes behind the outperformance.

Source: Business News Australia

Thursday, 23 July 2026

Wednesday, 22 July 2026

Tuesday, 21 July 2026

Market-structure

A trading signal that used to decay over five to seven years now loses half its edge in eighteen months, and AI crowding is the mechanism

An NYU study of nearly one million institutional fund holdings finds portfolios converging as AI adoption spreads, while separate tests show AI trading models beating human accuracy but running 20 to 40 percent daily volatility against a 7 to 15 percent target, evidence that AI is compressing alpha decay faster than it is improving risk discipline.

Source: Bloomberg

Monday, 20 July 2026

Sunday, 19 July 2026

Thursday, 16 July 2026

Research

Why asking AI trading agents beats telling them what to do

The most useful practical result from a year of agent research is boring: instruction quality dominates raw model capability. For finance teams, the prompt is now a first-class engineering artefact.

Source: FinTech Global

Retail

Kraken unveils an AI-powered crypto trading platform stack

Kraken becomes the latest retail-facing venue to bundle AI tooling into its trading UI, alongside an API partner program. The pattern is now standard, and the regulatory question is starting to catch up.

Source: Blockchain Council

Wednesday, 15 July 2026

Institutional

Singapore's Longbridge bets on AI agents to transform retail investing

Singapore-headquartered Longbridge Securities has announced a multi-agent AI investing platform aimed at the affluent-retail tier across Southeast Asia, targeting a segment that neither traditional wealth management nor discount brokers currently serve well.

Source: finews.asia

Tuesday, 14 July 2026

Regulation

AI trading assistants test regulatory readiness for the CLARITY Act

The US CLARITY Act's push toward tokenised assets and 24/7 markets is running headlong into a first-order question: what happens when the counterparty on the other side is an autonomous LLM agent operating without a human in the loop?

Source: Yahoo Finance UK

Sunday, 12 July 2026