The SEC Is Investigating Situational Awareness. The Real Story Is the Leverage, Not the Genius

TL;DR: Situational Awareness, the hedge fund run by former OpenAI researcher Leopold Aschenbrenner, peaked near $45 billion in assets under management in early July 2026, then lost roughly 67% of that value in a month, collapsing to somewhere around $8 billion to $10 billion after margin calls forced a fire sale of leveraged AI and chip bets. The SEC has now subpoenaed Goldman Sachs, JPMorgan Chase, Citigroup, and Bank of America over their lending and margin practices with the fund, according to reporting from Disruption Banking. Be direct with yourself about what we know: a subpoena is not a finding of fraud, and nobody has charged Aschenbrenner or his fund with wrongdoing. Here's my read on why this happened anyway, and why I think it happens again with the next AI genius who convinces banks he can't be wrong.

What Actually Happened at Situational Awareness

You need the timeline first, because the speed here is the whole story. Leopold Aschenbrenner launched Situational Awareness LP in July 2024 with about $225 million in seed capital from Patrick Collison, John Collison, Nat Friedman, and Daniel Gross, according to CNBC's reporting. That's a serious but unremarkable launch for a well-connected former OpenAI safety researcher who had just published a widely read essay series on AI timelines.

What happened next is where I start asking questions. Regulatory filings show the fund's reported 13F assets under management went from $552 million at the end of 2025 to $13.7 billion in the first quarter of 2026, a jump of roughly 2,396%, then to $20.2 billion by the second quarter, per data compiled by 13Finsight and 13F.info and reported by Bloomberg through the Economic Times. Add leverage on top of that reported equity base and you get to the $45 billion peak AUM figure CNBC and other outlets cited in early July. I'm not aware of any track record that justifies going from half a billion to tens of billions in six months. That's not performance. That's inflows chasing a narrative, with leverage magnifying whatever the narrative touched.

The fund's book was concentrated in AI and chip names: long positions in SK Hynix and CoreWeave, short positions including Adobe, all wagers mapping directly onto Aschenbrenner's public thesis that AI capability is compounding faster than markets price in. When SK Hynix and the broader AI-infrastructure trade stalled in July 2026, the fund didn't just lose money on the long book. It got margin calls. Reported leverage ran as high as 400%, meaning for every dollar of investor capital, the fund controlled roughly four dollars of market exposure. At that ratio, a 15% to 20% move against you wipes out the equity cushion and triggers forced selling.

Citadel, Ken Griffin's firm, reportedly bought the bulk of Situational Awareness's unwound public portfolio in a single overnight transaction, according to Business Insider's coverage cited alongside CNBC's reporting. That's the fire-sale mechanic in its purest form: a fund with no time left sells to whoever will buy, at whatever price the buyer names. Citadel didn't have to negotiate. It had liquidity and Situational Awareness didn't.

The SEC Investigation's Real Target Isn't Just the Fund

Here's the part I find more interesting than the collapse itself. The SEC didn't just open a file on Situational Awareness. It subpoenaed four of the largest banks in the country, Goldman Sachs, JPMorgan Chase, Citigroup, and Bank of America, over their trades, leverage arrangements, and internal communications with the fund's lenders, per Disruption Banking's account of the Reuters reporting.

I want to repeat the caveat because it matters: a subpoena means the SEC wants documents and testimony, not that it has concluded any bank or person broke the law. Investigations end in no action all the time. But regulators going straight to the prime brokers, not just the fund manager, tells you where the actual risk sits. A fund with $500 million in reported equity two years earlier does not get to $45 billion in exposure without banks competing to hand it leverage. Somebody at each of those four institutions signed off on financing terms that let a young fund run 4-to-1 leverage on a thesis about chip demand and AI capability curves. That's a lending decision, made by risk committees at sophisticated credit shops, and it's my opinion, not documented fact, that the bank-focused subpoenas are the real tell. Subpoenaing four lending desks means asking whether the pressure that made banks race to lend to Archegos in 2021 made them race to lend to Aschenbrenner in 2026.

The Archegos Parallel, With the Numbers That Actually Match

I've heard people wave off the Archegos comparison as lazy pattern-matching. I don't think it is; the mechanics line up too closely. Bill Hwang's Archegos Capital Management used total return swaps to build roughly 6x leverage on a concentrated set of stocks, including ViacomCBS, without disclosing the underlying positions to any single bank, because each bank only saw its own slice of the exposure. When those positions cratered in March 2021, the unwind cost banks more than $10 billion combined. Credit Suisse alone lost $5.5 billion, in large part because its own risk staff had proposed dynamic margining changes weeks before the collapse and the firm didn't act fast enough, according to the Credit Suisse Special Committee's own post-mortem report. Bill Hwang was convicted of fraud and racketeering in 2024.

Situational Awareness didn't use total return swaps in the same structure as Archegos, as far as public reporting shows, but the underlying failure mode is identical: concentrated bets on a handful of correlated names, leverage supplied by multiple banks with only partial visibility into total exposure, and a stall that forced a fire sale instead of an orderly exit. Swap semiconductor stocks for media stocks, swap four subpoenaed banks for Credit Suisse and Nomura absorbing losses, and you're reading the same case study five years later.

Metric	Archegos Capital (2021)	Situational Awareness (2026)
Founder	Bill Hwang	Leopold Aschenbrenner
Reported peak leverage	~6x via total return swaps	Up to ~400% (roughly 4x) reported
Peak exposure / AUM	~$20B+ in swap exposure on ~$10B equity	~$45B AUM peak (some estimates $30B+)
Concentration	Handful of names, incl. ViacomCBS	AI/chip names: SK Hynix, CoreWeave, Adobe (short)
Bank losses / exposure	$10B+ combined, Credit Suisse alone lost $5.5B	Not yet quantified, SEC probing bank exposure
Unwind mechanism	Block trades dumped by banks, no orderly exit	Overnight fire sale, bulk bought by Citadel
Regulatory outcome	Hwang convicted of fraud, 2024	SEC investigation open, no charges filed

My Contrarian Thesis: Genius Doesn't Fix a Broken Risk Model

Here's my actual argument, and it's not that Aschenbrenner is a fraud. I have no evidence for that. My argument is that AI-conviction hedge funds are structurally more likely to blow up this way than a generalist multi-strategy fund, and the founder's intelligence has nothing to do with fixing that structural problem.

Think about why banks extended Situational Awareness that much leverage that fast. Aschenbrenner wrote "Situational Awareness: The Decade Ahead," a widely circulated essay arguing AI capability is compounding faster than most people believe. That essay built him a reputation as someone who sees around corners on the biggest macro trend in markets right now. When a fund manager with that reputation asks for leverage, prime brokerage desks aren't underwriting a balance sheet. They're underwriting a narrative. That's the same mistake Credit Suisse's relationship managers made with Bill Hwang: treating a compelling story as a substitute for stress-testing what happens when the thesis is right on direction but wrong on timing.

That's the trap specific to AI-conviction funds. The AI capability story might be completely correct over five years and still cost you everything if you're leveraged 4x and the market corrects for two months. Timing risk and thesis risk are different risks, and leverage only punishes timing. Aschenbrenner could be exactly right that AI infrastructure demand triples by 2030 and still have been forced to liquidate in July 2026 because SK Hynix and CoreWeave stalled for a few weeks against a book that couldn't survive a 15% drawdown. A fund with real conviction and no leverage rides out that stall. A fund with 4x leverage doesn't get the chance.

I'd also point out the asymmetry in how fast capital showed up. Going from $552 million to $20.2 billion in reported 13F assets in two quarters isn't just leverage doing work. It's investors and prime brokers racing toward the same narrative at once, the crowding dynamic that makes unwinds violent. When everyone's long the same handful of names, there's no natural buyer when the selling starts. Citadel buying the whole unwound book overnight isn't a sign the system worked. It's a sign there was exactly one buyer with the balance sheet to do it.

What This Means If You're Considering Any Concentrated Hedge Fund Bet

I'll give you the checklist I'd use if a fund pitched me on a single dominant macro thesis, AI or otherwise.

First, ask about leverage explicitly, not performance. A 13F filing shows you long equity positions, nothing about swaps, options, margin borrowing, or synthetic exposure. Situational Awareness's public numbers looked like a fund up 80% on the year even after the July drawdown, according to reporting citing Bloomberg data. That headline number tells you nothing about whether the fund was 4x levered to get there. Ask the manager directly what the gross exposure to net asset value ratio is, and what happens to that ratio in a 20% adverse move.

Second, treat rapid AUM growth as a risk flag, not a credibility signal. A fund going from $552 million to $20 billion in two quarters isn't a track record. It's a stress test waiting to happen, because the risk infrastructure built for a $500 million fund rarely scales cleanly to a $20 billion one in six months.

Third, ask who the prime brokers are and whether they're the same banks providing leverage across multiple funds riding the same thesis. If Goldman, JPMorgan, Citigroup, and Bank of America extend credit against similar AI-infrastructure collateral to multiple funds, you have correlated risk hiding behind supposedly diversified lending relationships. That's precisely the blind spot that let Archegos borrow from Credit Suisse, Nomura, Morgan Stanley, and UBS simultaneously without any one bank seeing the full picture.

Fourth, separate the thesis from the structure. I can believe the AI capability story and still refuse to back a fund expressing that belief through 4x leverage on five correlated names. Conflating those decisions is how investors end up defending a manager's intelligence instead of his risk controls.

None of this means Situational Awareness did anything illegal, and none of it means Aschenbrenner's AI thesis is wrong. The SEC investigation could close with no findings. But you don't need to wait for that conclusion. The leverage numbers and the speed of the unwind are public, and they tell you enough about the structure regardless of how the legal question resolves.

Frequently Asked Questions

Has the SEC accused Leopold Aschenbrenner or Situational Awareness of wrongdoing?

No. As of late August 2026, the SEC has subpoenaed banks including Goldman Sachs, JPMorgan Chase, Citigroup, and Bank of America for records on their lending and trading relationship with Situational Awareness. That's an investigative step, not a charge. No enforcement action or formal fraud allegation has been filed against Aschenbrenner or the fund, and investigations like this often close with no finding of wrongdoing.

How much money did Situational Awareness lose in the July 2026 collapse?

Reported figures vary by source, but the fund is widely reported to have peaked near $45 billion in assets under management in early July 2026 before falling approximately 67% within a month, to somewhere between $8 billion and $10 billion, following margin calls tied to leveraged bets on AI and chip stocks. Some reports cite a lower peak closer to $30 billion-plus. The fund reportedly remained up roughly 80% for the year even after the drawdown, according to data cited by Bloomberg.

Is the Archegos comparison actually fair, or is it just a scary headline?

I think it's fair on the mechanics, not necessarily on the legal outcome. Both funds used significant leverage, roughly 6x for Archegos via total return swaps and reportedly up to 4x for Situational Awareness, to build concentrated positions in a small number of correlated names. Both funds' exposure was spread across multiple banks that couldn't individually see total leverage. Both unwound through forced, disorderly sales. Bill Hwang was later convicted of fraud, but that outcome isn't guaranteed here, and nothing in the current record establishes that Aschenbrenner engaged in the kind of deception Hwang was convicted for.

What should an accredited investor do differently before allocating to a concentrated thesis-driven hedge fund?

Ask for gross exposure to net asset value ratios directly, not just performance figures. Treat rapid AUM growth, like Situational Awareness's move from $552 million to over $20 billion in two quarters, as a risk flag rather than proof of skill. Find out which prime brokers extend leverage and whether those banks finance other funds riding an identical thesis, since that correlation is invisible in any single fund's disclosures. Separate your view of the investment thesis from your view of the fund's risk structure. You can believe the AI story and still decline the leverage.

Author Disclosure: Jeff Barnes, MBA has no personal position in any company, fund, or platform named in this article. Angel Investors Network has no current commercial relationship with any party mentioned. AIN provides marketing and education services, not investment advice. Past performance does not guarantee future results. All investments involve risk, including loss of principal.