A startup in New York is making a provocative claim: its hedge fund lets artificial intelligence handle every investment decision, no questions asked. No portfolio manager reviewing trades before they execute. No oversight committee second-guessing the algorithms. Just AI, start to finish.
Standard Signal, which emerged from Y Combinator's cohort earlier this year, bills itself as "the first hedge fund that researches and executes trades purely with AI." The founding pitch promises AI agents capable of discovering "new fundamental truths before humans can"—a line that would sound ambitious in any era, let alone one where ChatGPT still occasionally hallucinates stock tickers.
Whether any of this holds up depends on three things: how you define "first," what you mean by "purely AI," and whether you've been paying attention to the last decade of similar promises.
The Latest Wave
Standard Signal launched in 2026 under Michael Royzen, who previously built Phind, the YC-backed AI search engine aimed at developers. The fund's website describes a system that "trains and deploys frontier financial LLM models with full trading autonomy within strict risk parameters"—claims that remain unverified publicly and represent the company's marketing positioning. It targets a market-neutral strategy with a Sharpe ratio north of 2.0 and sets the entry point at $100,000.
According to Y Combinator's directory from Spring 2026, the team consisted of one person.
Standard Signal has company in this race, though claiming pole position gets tricky. Lumenai Innovation Fund announced in April that it "believed to be the first institutional hedge fund built on an agentic AI architecture," with operations slated to begin around June 1. Phasic debuted its MLX Fund in February, marketing it as an agentic AI-managed vehicle focused on S&P 500 stocks. RAM Active Investments has been building what Risk.net called an "agentic multi-strategy" operation, though the firm's own May 2025 materials described it more cautiously as a multi-strategy expansion.
Even the ETF world got in on the action. FINQ, based in Israel, launched two U.S.-listed funds in February—tickers AIUP and AINT—which Reuters reported as what the company claimed were the first SEC-approved ETFs "managed solely by AI."
A Short History of 'First'
Here's the problem with claiming to be first: someone usually beat you to it.
Aidyia, a Hong Kong fund, launched in January 2016 with Wired declaring that "AI makes all trades with no human intervention." Sentient Investment Management spun out of Sentient Technologies the same year with similar claims. It shuttered in 2018.
Those experiments unfolded in a different technological landscape—pre-transformer architectures, pre-GPT, before anyone was using the word "agentic" in pitch decks. Whether they technically count as "first" hinges on how you score autonomy and what level of human involvement disqualifies a system from being purely algorithmic. The historical record, at minimum, complicates the narrative.
What's different now is the technology stack. Large language models can parse earnings calls, generate backtesting code, and handle multi-step reasoning chains in ways that weren't possible eight years ago. The infrastructure exists. The question remains whether it works—and whether it works well enough, consistently enough, to justify handing over capital.
How the Industry Actually Uses AI
While a handful of startups chase full autonomy, the established players are taking a different path. They're deploying AI, yes, but mostly in the unglamorous back and middle office first.
Goldman Sachs started piloting Anthropic's Claude for trade accounting and compliance checks, according to February trade press coverage. Citi said in April it was "moving into agentic AI," per Axios. JPMorgan's Jamie Dimon told investors on May 21 that the bank would hire more AI specialists and fewer traditional bankers—a strategic direction that signals where leadership thinks the leverage points are.
Man Group, the publicly traded alternative manager, published research last fall on how it uses agentic frameworks for research automation: pulling documents, parsing earnings transcripts, synthesizing information across sources. But, crucially, with humans still in the loop. The writeup was candid about limitations: AI agents handle defined tasks well but stumble on edge cases and demand rigorous data discipline.
Two Sigma and AQR have woven machine learning into their processes for years. What's shifted in 2026 is the emphasis on agentic systems—AI that plans, executes, and iterates without constant human prompting—rather than static models that need periodic retraining.
Numerai, the market-neutral fund that's crowdsourced machine learning signals since 2016, continues to run an AI-native pipeline, though it shares little performance data publicly.
Augmentation, not replacement. That's been the playbook so far.
The Regulatory Gray Zone
Regulators don't have finalized rules specifically for AI trading yet, but they're circling.
In the U.S., the SEC proposed a rule in 2023 targeting conflicts of interest from predictive analytics and AI, then withdrew it on June 12, 2025. A revised version was reportedly under consideration as of last spring. As of May, no finalized "AI conflicts" rule exists, leaving firms to navigate existing fiduciary duties, Reg BI, books-and-records mandates, and model risk governance expectations—a patchwork that's principles-based, not prescriptive.
FINRA's 2026 oversight report, released in December, flagged generative AI governance as a priority. Broker-dealers should expect exam questions on AI due diligence, testing, explainability, supervision, and vendor risk.
The CFTC's Electronic Trading Risk Principles, finalized in 2021, remain the baseline framework. The agency sought public comment on AI in January 2024 and has delivered speeches emphasizing controls to prevent disorderly trading and monitoring for manipulation.
Europe's AI Act implementation timelines shifted in May when negotiators reached a provisional agreement under the "Digital Omnibus." Earlier deadlines—some as soon as August 2—were pushed out, with high-risk stand-alone systems now facing compliance obligations by December 2, 2027, and high-risk embedded systems by August 2, 2028, pending finalization. The European Commission released draft guidance on high-risk classification on May 20.
ESMA issued a supervisory briefing on algorithmic trading in February, emphasizing governance, testing, and outsourcing clarity for AI-driven trading on EU venues. The briefing reinforced expectations around pre-trade controls, kill switches, and audit logs under MiFID RTS 6.
In the UK, the Prudential Regulation Authority's Model Risk Management principles—effective since May 2024—apply to banks and set expectations for model inventory, validation, and board oversight. The FCA published a multi-firm review on algorithmic trading controls last August, with updated observations in March highlighting weaknesses in firms' compliance. On May 15, the FCA, Bank of England, and HM Treasury issued a joint statement on frontier AI and resilience, reinforcing supervisory expectations without introducing new mandates.
The regulatory landscape is evolving, not settled. Firms can deploy AI agents, but they need to show robust governance, explainability, and risk controls. There's flexibility, but also scrutiny.
Academic Research Meets Market Reality
Academic work on AI trading has proliferated, but the quality varies wildly.
A survey published on arXiv in May audited 77 studies on LLM-based trading strategies and found severe reproducibility issues. Many papers used inconsistent protocols, omitted execution frictions like slippage and transaction costs, and failed to provide enough detail for independent replication. The authors concluded that "poor protocol comparability and low reproducibility" plague the field—a polite way of saying: don't trust the backtests.
Earlier research from 2023, including a widely cited paper by Lopez-Lira and Tang, showed that LLMs could extract predictive signals from financial news over short horizons. Later replications suggested the edge decayed as adoption increased and markets internalized the signals. Alpha, in other words, doesn't sit still.
A separate CEPR discussion paper, revised in February, examined systemic risk implications. The authors modeled agent investors—both Q-learning algorithms and LLM-based agents—and found they could amplify run dynamics during stress. As AI penetration grows, coordination and herding risks rise, creating potential instability that regulators will need to monitor. It's the 2010 flash crash concern, updated for the LLM era.
Man Group's practitioner perspective from late 2025 struck a more measured tone: agentic systems work well for defined research tasks—pulling data, summarizing documents, generating code—but struggle with ambiguous edge cases and require continuous human oversight. Data discipline matters. Validation matters. Blind reliance on model output? Not recommended.
What Happens Next
The hedge fund industry reached $5.22 trillion in assets under management in the first quarter, according to HFR's April 23 update, up from $5.15 trillion at year-end 2025. New fund launches accelerated through last year, with 427 new funds through the third quarter, per HFR's January report. Liquidations remain near historic lows.
AI-native funds represent a rounding error in that capital base. Standard Signal's $100,000 minimum suggests it's targeting high-net-worth individuals and smaller family offices, not the institutional allocators who move the industry. Lumenai and Phasic are similarly early-stage. None have published audited, time-weighted performance records covering a full market cycle—because none of them have been around long enough to experience one.
IMARC pegged the algorithmic trading market at $18.8 billion in 2025, with growth projected through 2034. That's a vendor forecast with methodology that varies, but the direction is clear: automation is accelerating.
Mercer's May 21 report on AI in asset management found that firms have moved past experimentation. AI now augments human investment processes at scale. KPMG's Q1 pulse survey showed rising pilots of AI agents across asset management and private equity operations. But augmentation isn't autonomy. The industry is adopting AI to make humans more efficient, not to replace them wholesale. Not yet, anyway.
The real test isn't whether AI can make trades without human intervention—that's technically feasible today. The test is whether it can generate alpha consistently, explain its reasoning to investors and regulators, and navigate the inevitable market dislocation when something breaks. The algorithms of 2016 couldn't. Whether the models of 2026 can—and whether anyone will trust them enough to find out—remains an open question.
For now, Standard Signal and its peers are placing a bet: that the technology has finally caught up to the promise. The market will render its verdict in time. It always does.
