SEC subpoenas banks over AI hedge fund collapse
The SEC has subpoenaed Goldman Sachs, JPMorgan, Citigroup, and Bank of America for records tied to the Situational Awareness AI hedge fund collapse. This dated Tracker entry records which banks were named, what the probe does and does not cover, and what to watch next as the inquiry unfolds.
- Platform
- SEC
- Change category
- policy
- Change type
- policy shift
Tracker status, Aug. 26, 2026: the SEC has reportedly subpoenaed Goldman Sachs, JPMorgan, Citigroup, and Bank of America for records tied to the collapse of Situational Awareness, the AI-linked hedge fund whose July unwind turned into one of the cleaner tests of how regulators will treat AI-adjacent market risk. CNBC, carrying Reuters-confirmed reporting on Aug. 25, named the four banks and described an early-stage inquiry focused on trades that triggered margin calls, leverage use, communications with lenders, and bank record preservation demands.[1]
That is the operative update. The subpoena reporting does not say the SEC is directly investigating the fund’s AI models. It does not establish that the banks or the fund did anything wrong. It does mean the collapse has moved from market rumor and investor damage control into a regulatory record, with named institutions now reportedly asked to preserve and produce documents.

What the SEC reportedly asked banks for
The cleanest available formulation is narrow: the SEC is seeking bank records around the fund’s margin-call event chain. CNBC’s Aug. 25 report, citing Reuters confirmation, said subpoenas went to Goldman Sachs, JPMorgan, Citigroup, and Bank of America.[1] Quartz’s Aug. 25 coverage described the same inquiry as focused on when Situational Awareness executed trades that triggered margin calls, how much leverage the fund used, lender communications, and demands that banks preserve records.[2]
That scope matters because it places the first public regulatory emphasis on financing, disclosure, and communications rather than on whether an AI system picked the wrong stocks. Hedgeweek’s account of the Reuters-confirmed reporting similarly framed the subpoenas around leverage and margin-call records, not a model audit.[3]
| Reported subpoena area | Why it matters |
|---|---|
| Trades that triggered margin calls | Creates a dated sequence for when stress moved from portfolio losses to lender action. |
| Leverage use | Tests what was represented to investors, lenders, and counterparties against the financing record. |
| Communications with lenders | Shows what the fund and banks said to each other while positions were being financed, pressured, or unwound. |
| Record-preservation demands | Turns bank-side materials into regulatory evidence even before any enforcement outcome is known. |
The earliest public report of the SEC probe was attributed to The New York Times on Aug. 24 and then carried through accessible secondary coverage, including TechCrunch.[4] The next day’s Reuters-confirmed reporting is the more useful marker for this Tracker entry because it named the banks and clarified the reported subpoena target set.[1]
Why those records matter: the July unwind
The fund’s collapse context is relevant only to the extent it explains why bank records would matter. Situational Awareness reportedly entered the summer with a highly concentrated public-equity book. Quartz and Cryptopolitan, citing the fund’s June 30 13F, reported about $20.24 billion in U.S. holdings, with SanDisk and Micron together representing more than 56% of that book.[2][5]
The July sequence then compressed quickly. Reports describe a July 24 investor letter, a late-July drawdown, and a July 30 sale of the fund’s public-equity portfolio to Citadel at a reported discount of about 10%.[6][7] CNBC’s July 31 coverage, citing The Wall Street Journal, said the fund had fallen about 67% in July and described assets dropping from roughly $45 billion to about $10 billion.[7]
Those figures should not be cleaned up into a single neat AUM number. CNBC and Quartz carried the roughly $45 billion-to-$10 billion framing; Axios described the vehicle around a roughly $20 billion scale; Cryptopolitan, citing Wall Street Journal reporting, referenced a reported $100 billion peak.[2][5][6][7] The defensible reading is that the drawdown was large, rapid, and sufficiently tied to financing pressure that regulators are now reportedly asking banks for the paper trail.

The leverage number also belongs in the “reported claim” column. Quartz described leverage as reportedly reaching up to 400%.[2] That is not the same as a regulatory finding. It is, however, exactly the sort of claim that makes margin-call timing, lender communications, and disclosure records consequential.
What this is not: an AI model case, at least on the public record
The shorthand “AI hedge fund collapse” is easy and partly unavoidable. Situational Awareness was publicly framed around AI, and the size of the unwind makes that label sticky. But the subpoena reporting available so far points somewhere more prosaic: leverage, trades, lender communications, and preservation of bank records.
That distinction is not cosmetic. A probe into model design would raise one set of questions: training data, explainability, human supervision, testing, and model governance. A probe into leverage disclosure and margin lending raises another: whether investors and counterparties were told enough, when stress became visible, who knew what, and how the financing stack behaved during the unwind.
The fund’s own public posture has also been cooperation, not combat. CNBC and Quartz reported that Situational Awareness said scrutiny of high-profile drawdowns is expected and that it is fully cooperating.[1][2] That statement does not resolve the inquiry; it simply belongs in the record alongside the subpoenas and the absence of any public allegation of wrongdoing.
The Atkins-era SEC context
The subpoena reporting lands inside an SEC posture that has been described less as broad AI rulemaking-by-headline and more as fraud-first enforcement. The agency’s FY2025 enforcement results, released Apr. 7, 2026, reported 456 enforcement actions and said about two-thirds of standalone actions charged individuals, up 27% year over year.[8]
The agency also created a Cyber and Emerging Technologies Unit in February 2025, a structure that keeps AI, cyber, crypto, and other technology-linked misconduct inside an enforcement frame rather than treating every emerging-technology failure as a technology-policy case.[9] In a Feb. 3, 2026 Division of Investment Management speech, SEC official Daly addressed AI in investment management through obligations, disclosures, and governance rather than through a blanket claim that AI systems are inherently unlawful.[10]

That is the useful lens for this case. If an AI-branded investment product fails, the first regulatory question may not be whether AI “caused” the loss. It may be whether the humans and institutions around the product accurately described leverage, liquidity, risk controls, and decision responsibility.
Why this belongs in an AI regulation tracker for marketers
This is not an ad-platform case. No source in the current record says the SEC subpoenas have anything to do with paid media, ad targeting, or marketing automation. The connection for Signal & Convert readers is structural: AI claims move fastest in markets before they harden into compliance language.
For performance marketers, the practical signal is how regulators sort an AI-adjacent failure. If the public record keeps centering on disclosure, financing, and communications, that says more about risk language than about model panic. It is closer to the discipline used in our AI ad policy tracker: identify the obligation, separate reported facts from claims, and do not turn every AI-linked incident into a platform-wide rule change.
The same reading habit matters when a market event affects vendors or media infrastructure. In the Trade Desk stock-drop benchmark, the question was what actually changed for advertisers versus what only changed in market sentiment. Here, the equivalent distinction is what changed in the regulatory record versus what remains narrative pressure around AI.
What to watch next, dated Aug. 26, 2026
- Whether the inquiry closes without action or becomes a formal enforcement action. Subpoenas are investigative tools, not findings.
- Whether additional banks, prime brokers, lenders, or counterparties are named in credible reporting.
- Whether the public scope remains concentrated on leverage, margin calls, and communications, or expands toward investor statements about AI strategy and risk controls.
- Whether future AI-adjacent financial failures are framed by the SEC around disclosure and fraud rather than the underlying AI systems.
- Whether bank-side margin-lending scrutiny broadens beyond this fund, which would make the event more than a single-fund collapse story.
References
- Reuters-confirmed reporting on SEC subpoenas to Goldman Sachs, JPMorgan, Citigroup, and Bank of America, CNBC, Aug. 25, 2026.
- Coverage of SEC subpoena scope and Situational Awareness 13F concentration, Quartz, Aug. 25, 2026.
- Coverage of Reuters-confirmed SEC subpoenas tied to Situational Awareness, Hedgeweek, Aug. 2026.
- Coverage attributing the first report of the SEC probe to The New York Times, TechCrunch, Aug. 24, 2026.
- Coverage of Situational Awareness drawdown, AUM claims, and June 30 13F concentration, Cryptopolitan, Aug. 25, 2026.
- Coverage of the July 30 fire sale of the public-equity book to Citadel, Axios, July 30, 2026.
- Coverage of Situational Awareness July losses and reported asset decline, CNBC, July 31, 2026.
- SEC Announces Enforcement Results for Fiscal Year 2025, U.S. Securities and Exchange Commission, Apr. 7, 2026.
- SEC Announces Cyber and Emerging Technologies Unit, U.S. Securities and Exchange Commission, February 2025.
- Division of Investment Management AI speech, U.S. Securities and Exchange Commission, Feb. 3, 2026.
Primary source: https://www.sec.gov/newsroom