Over two days at the end of August, subpoenas went out to several of Wall Street’s largest banks. The recipient of the questions was not a bank itself, but a hedge fund: Situational Awareness, the AI-driven trading operation that once ranked among the most celebrated funds in the industry. The Securities and Exchange Commission is investigating the fund’s near-collapse, according to people familiar with the matter, and it has asked the banks that serviced it for trading records, margin agreements, and communications.
The investigation closes a dramatic arc. Situational Awareness launched with a pitch that captured the imagination of the AI era: an entirely machine-driven fund, built by researchers rather than traders, that would find patterns in markets no human could see. It raised money at valuations that made it one of the fastest-growing funds in history, and its early returns attracted a list of investors that read like a who’s who of technology finance.
The collapse, when it came, was fast. The fund’s models, according to people familiar with its operations, were designed to exploit short-lived inefficiencies across thousands of markets, a strategy that works while the models are right and fails violently when they are wrong. In the weeks before the near-collapse, the models began to fail in ways the risk systems did not anticipate, and the losses compounded before anyone could unwind the positions.
The exact size of the losses has not been disclosed. What is known is that the fund’s counterparties, the banks that lent it money and cleared its trades, moved to protect themselves, demanding more collateral and cutting credit lines as the losses mounted. The subpoenas are designed to determine what the banks knew, when they knew it, and whether the fund’s disclosures to investors matched what was happening in its trading book.
The case is the first major regulatory test of AI-driven investing. Machine learning has been used in markets for years, but mostly in narrow corners: market making, execution, and the slow accumulation of statistical edges. Situational Awareness represented something bolder, a fund that treated the entire market as a single optimization problem, and its failure has become the cautionary tale for that ambition.
The SEC’s interest extends beyond the fund itself. The investigation is examining whether the banks’ risk controls were adequate for the kind of debt-financed, machine-driven positions the fund took, and whether the industry’s infrastructure, built for human traders with human reaction times, is equipped for models that can double down in milliseconds. The answers could shape how banks deal with AI funds for years.
The near-collapse also feeds a broader debate about concentration. The fund’s positions, by some accounts, accounted for a meaningful share of trading volume in several markets, which means its failure risked cascading into the wider financial system. Regulators have worried publicly about exactly this scenario, a single AI fund big enough to move markets and fragile enough to fail at speed.
For investors, the lesson is being priced in real time. Funds that advertise AI strategies are facing renewed scrutiny from allocators, who are asking harder questions about risk models and drawdown protection. Some of the money that flowed to Situational Awareness is expected to return to more conventional funds, at least until the industry can show that its machines can fail gracefully.
The fund’s name, borrowed from a term in military and security doctrine, captured the ambition. Situational awareness, in that tradition, means knowing what is happening around you before it happens to you, and the fund’s founders argued that AI had finally made that possible at market scale. The pitch attracted investors who saw the fund as a hedge against the inefficiencies of human-driven markets, and its early results appeared to confirm the thesis. What the models could not do, the collapse showed, was maintain that awareness when the environment itself changed.
The investigation will focus on the mechanics of the failure. Regulators want to know how the fund’s borrowing was built, how its risk models measured the probability of the events that actually occurred, and whether its disclosures to investors painted an accurate picture of the strategy. The banks, which provided the credit, will face questions about their own risk management: whether their systems, designed to flag human trading patterns, could see what a machine-driven book was doing in real time.
The broader implication for the industry is uncomfortable. If the collapse of a single AI fund can attract subpoenas across Wall Street within days, the regulatory environment for machine-driven investing is about to change. Funds that rely on opaque models and rapid-fire trading will face demands for explainability, and banks that service them will face demands for oversight, whether the technology cooperates or not. The era of AI trading on trust is over; the era of AI trading under audit has begun.
The investigation is at an early stage. The banks have said they are cooperating, and the SEC has given no indication of when it might conclude. What is already clear is that the era of AI hedge funds has entered a new phase: the marketing is over, the audit has begun, and the machines are no longer the only ones being tested.
