The AI Hedge Fund Everyone Was Copying Last Filed a Book That Was 62% Puts

For two years, a group of retail investors followed the quarterly SEC filings of Situational Awareness, the AI-themed hedge fund founded by former OpenAI researcher Leopold Aschenbrenner, looking for stock ideas. In July 2026 the fund told investors its unaudited net return for the month was −67%, and sold the bulk of its public stock book to Citadel under margin pressure.

So I pulled the filings. The fund’s most recent Form 13F, filed on May 18, 2026 for the quarter ended March 31, reports $13,676,657,577 across 42 entries. Sixty-two percent of that value sits in put options, eleven of them, on semiconductor and AI infrastructure names. Only 28% is ordinary long stock.

That is not the portfolio the news coverage described, and it is not the portfolio anyone was copying. The gap between those two statements is the most useful thing a DIY investor can take from this episode, and it has nothing to do with whether AI is a bubble.

What the last filed portfolio actually showed

Situational Awareness LP has filed six 13Fs since it began reporting. The most recent covers March 31, 2026. Here is how its reported value breaks down by row type.

The eight largest put lines are the VanEck Semiconductor ETF ($2,042,716,860), NVIDIA ($1,568,257,120), Oracle ($1,072,873,230), Broadcom ($1,006,247,961), AMD ($969,160,863), Micron ($583,686,168), Taiwan Semiconductor ($535,110,030), and ASML ($494,122,503). Together they are $8,272,174,735, or 60.5% of everything the fund reported.

The long stock is real too, and it is the part that matches the public story: Bloom Energy ($878.7M), SanDisk ($724.4M), CoreWeave ($556.1M), IREN ($401.0M), Core Scientific ($389.1M), and Applied Digital ($320.0M), alongside a set of smaller bitcoin miners and power names. That is the AI infrastructure thesis on the page. It is also 28% of the filing.

A reader who scanned this filing for tickers to buy would have taken away six or eight AI infrastructure names and skipped the majority of the book, because the majority of the book was options and options are harder to read. The next three sections are about why.

A two billion dollar line that never cost two billion dollars

The VanEck Semiconductor ETF put line reports $2,042,716,860 on 5,327,900 underlying shares. Divide one by the other and you get $383.40, which is the price of a share of that ETF on March 31, 2026.

Form 13F reports option positions at the market value of the underlying shares. The fund did not spend two billion dollars on puts. It held puts covering two billion dollars of stock, and the premium it paid appears nowhere in the filing.

You do not have to take an outside price source’s word for this. The filing proves it against itself. Eleven issuers appear on both an option line and a share line, and in all eleven the implied price per share agrees to the cent:

This applies to every 13F. Aggregators rank a manager’s “top holdings” by reported value, and a modest options position will outrank a large stock position on that measure. The capital committed to those eleven put lines could have been a few hundred million dollars or a few billion. The form does not say.

The form has no column for long or short

It is tempting to read eleven put lines as a large bearish bet on semiconductors. That reading may well be right, and it is still an inference rather than something the filing states.

A put you buy pays off when the stock falls. A put you write pays off when the stock holds up, and it is a bullish position. Form 13F has one field, putCall, which records the contract type. It has no field recording whether the manager is long or short that contract.

The SEC’s own guidance makes the ambiguity narrower but does not remove it: managers are instructed not to report options they write, so a reported option is presumptively one the manager holds. That presumption rests on correct filing practice rather than on anything disclosed in the document. Strike prices and expiration dates are not collected at all, so even a confidently-read put tells you nothing about how far the stock has to fall, or by when.

The same filing shows why single-line readings are risky. Micron and Taiwan Semiconductor each appear as a put, a call, and a share position. Any story that begins “the fund was short Micron” has to account for the Micron calls sitting three rows above.

The positions that never appear

Press coverage of the unwind named SK Hynix, Nebius, and a short position in Adobe. None of the three is in the filing, and two of them could never have been.

• SK Hynix trades in Korea. The SEC instructs that shares trading on non-US exchanges are not reported on Form 13F.
• The Adobe short is invisible by rule. Managers are told not to include short positions, and not to net a short against a long in the same issuer.
• Nebius is the interesting one, because it is Nasdaq-listed and therefore reportable. It appears in none of the fund’s six filings. Either the position was built after March 31 or it was held in a form the filing does not capture.
• The Anthropic stake, reported in the press at roughly $5 billion, is a private holding. Only securities on the SEC’s Official List are reported, and private company equity is not on it.

Swaps, futures, cash, and ordinary bonds are absent for the same reason. A manager can hold economic exposure through a total return swap and disclose nothing. Two further gaps are worth knowing about: positions under 10,000 shares and under $200,000 may be omitted entirely, and a manager can ask the SEC for confidential treatment, filing the withheld positions separately as a 13F-CTR and keeping them out of public view for an initial three to twelve months.

Add these together and the honest description of a 13F is a partial census of one side of one asset class.

Forty-five days is a long time in a leveraged book

Managers have 45 days after quarter end to file. This one used 48 days, filing a March 31 snapshot on May 18. By the time the fund was selling stock to Citadel at the end of July, the most recent public document described a portfolio four months old. The filing covering June 30 was not due until August 14, after the unwind was over.

The lag matters more the faster a book moves, and this book moved quickly. Reported value went from $5.5 billion at the end of December to $13.7 billion at the end of March, with the put lines going from one small position to eleven worth 62% of the total.

Leverage is what turns a stale snapshot into a dangerous one. CNBC reported that traders pointed to reports the fund used as much as 400% leverage, a figure worth treating as a single unconfirmed claim rather than an established fact. A fund financed that way does not get to choose when it sells. Markus Brunnermeier and Lasse Pedersen described the mechanism. A shock to a borrower’s capital shrinks its capacity to absorb risk, which widens price swings. Wider swings raise lenders’ estimates of volatility, so margins go up. The borrower then has to sell into the same falling market. They call these the margin spiral and the loss spiral, and each one feeds the other.

The empirical version is unflattering for anyone hoping to shadow a levered manager. Studying hedge funds through the 2007 to 2009 crisis, Itzhak Ben-David, Francesco Franzoni, and Rabih Moussawi found funds sold about 29% of their aggregate equity portfolio in the third and fourth quarters of 2008, and that the selling concentrated in volatile and liquid stocks. Managers under margin pressure sell what moves easily, not what they have concluded is overvalued. A copier reading the resulting filing sees the residue of a financing problem and can easily mistake it for a view.

If you want the mechanics of margin calls and maintenance requirements applied to your own account rather than a hedge fund’s, that is the subject of Personal Leverage, which includes a margin stress test calculator.

What the research says about copying filings

The idea that you can free-ride on disclosed institutional holdings has been tested, and the results are worth knowing before you build a strategy on it.

John Griffin and Jin Xu examined hedge fund equity holdings directly and found hedge funds beat mutual funds at stock picking by 1.32% per year on a value-weighted basis, a result that was not statistically significant on an equal-weighted basis, with no evidence of an ability to time sectors or pick styles. Their own summary is that the study “raises serious questions about the perceived superior skill of hedge fund managers.” That is the ceiling on what copying can deliver before costs.

The more pointed finding concerns what managers are permitted to hide. Vikas Agarwal, Wei Jiang, Yuehua Tang, and Baozhong Yang studied positions withheld under confidential treatment and found they outperformed for up to twelve months after filing. George Aragon, Michael Hertzel, and Zhen Shi found the same pattern and showed managers are most likely to seek confidentiality for illiquid positions vulnerable to front-running. The positions with the best measured returns are the ones a copier cannot see.

The public record is also less fixed than it looks. Sean Cao, Zhi Da, Xin Daniel Jiang, and Baozhong Yang documented that 13F restatements are about as common as confidential filings but affect three times as many stocks, and concluded that some original holdings are misreported to obscure trading intentions.

Copycat strategies are not uniformly hopeless. Marno Verbeek and Yu Wang built hypothetical funds that duplicated disclosed holdings and found they generated performance comparable to their targets net of costs, with better results after the SEC tightened disclosure rules in 2004. Two caveats keep that from being a recommendation here: their targets were actively managed mutual funds rather than hedge funds, and matching a group that mostly trails the index is not the same as beating it.

When a 13F is worth reading

None of this makes the form useless. It is a genuine window into institutional behavior, and it is free.

Three uses hold up. Aggregate positioning across many managers is informative about crowding, which is a risk measure rather than a buy list, and the July momentum reversal is a reminder of what happens when crowded positions unwind together. Long-only managers running concentrated, low-turnover portfolios are the case where a quarterly snapshot most nearly describes the real book, since there are no shorts, little leverage, and few positions to miss. And tracking one company’s institutional ownership over time is a reasonable input to research on that company.

There is practitioner work claiming systematic 13F strategies can add several points a year, most visibly a Barclays quantitative research paper reporting outperformance from a conviction-and-consensus screen across selected managers. It is a sell-side working paper rather than peer-reviewed research, and it requires choosing the right managers in advance, which is the hard part of the problem restated.

What does not hold up is treating a leveraged long-short fund’s 13F as a portfolio you can reproduce. Roughly 70% of this fund’s reported value sat in instruments whose direction, cost, strike, and expiry were undisclosed, and its largest reported exposures by economic effect may have been ones the form never collected.

Work through what a copier could and could not have known:

How Summitward helps

The version of this problem that affects your own money is usually smaller and more tractable: knowing what you own, in aggregate, across accounts. Concentration and factor exposure are easy to underestimate when positions are spread across a 401(k), a brokerage account, and employer stock.

Frequently asked questions

What is a Form 13F?

A quarterly report that institutional investment managers with at least $100 million in qualifying US equity holdings must file with the SEC. It lists positions in securities on the SEC’s Official List, which is primarily US exchange-traded stocks, closed-end funds, ETFs, and certain convertible debt, equity options, and warrants. It is due within 45 days of quarter end.

Can I see a hedge fund’s short positions?

Not from a 13F. The SEC instructs managers not to report short positions, and not to subtract a short from a long position in the same issuer. Written options are excluded on the same basis. For a long-short fund, this means the form shows you one side of the strategy and gives no indication of the size of the other.

Why do the option values look so large?

Because they are reported at the market value of the underlying shares rather than the premium paid. A position controlling $2 billion of stock might have cost a small fraction of that. Ranking a manager’s holdings by reported value will therefore overstate options relative to stock.

Does a fund have to disclose everything it holds?

No. Beyond the categories the form excludes, managers may omit positions below 10,000 shares and $200,000 in value, and may request confidential treatment for specific positions, which keeps them out of the public filing for an initial three to twelve months. Research on those withheld positions finds they perform better than the disclosed ones.

Is there any way to copy a hedge fund accurately?

Not from public filings. You would need the short book, the derivatives, the financing terms, the private holdings, the cost basis, and the trades since quarter end, none of which are disclosed. Copying visible long positions from a hedged strategy can leave you holding one leg of a trade that was never meant to stand alone.

Did this fund’s AI thesis turn out to be wrong?

A single month does not settle that, and I take no position on it here. The fund reported a net gain of 80% for 2026 through the end of July even after a 67% monthly loss, which tells you the drawdown followed an extraordinary run rather than a straightforwardly wrong call. What the filings do show is a portfolio whose structure was far harder to read than its ticker list suggested.

Key takeaways

• Option lines on a 13F are notional, not premium. This filing proves it internally: eleven issuers carry the identical price per share on their option and share lines. Reported value measures exposure, and the money committed is undisclosed.
• The form cannot tell you long from short. There is no direction field, no strike, and no expiry. Short stock and written options are excluded outright, so a long-short fund’s filing is structurally one-sided.
• The best positions are often the ones you are not shown. Holdings withheld under confidential treatment have been found to outperform disclosed ones for up to a year, and managers withhold precisely the illiquid positions most vulnerable to being copied.
• A 45-day lag is fatal to a fast book. This fund roughly tripled its reported value in one quarter. Its March 31 snapshot was four months stale before anyone could read a newer one.
• Forced sellers sell what is liquid. Hedge funds under margin pressure in 2008 sold about 29% of their equity portfolios, concentrated in liquid names. A filing produced under that pressure records a financing constraint, and reads like a view.
• 13Fs are better for measuring crowding than picking stocks. Aggregate positioning, low-turnover long-only managers, and single-name ownership trends are the uses that survive contact with what the form leaves out.

Related guides

• Personal Leverage for margin mechanics, maintenance requirements, and a stress test on your own account.
• Conviction vs. Diversification on why researching a handful of stocks does not make them safer.
• Four Signs of a Bubble for a framework on the AI trade and a stress test of your exposure to it.
• Position Sizing on how large a speculative bet should be when you are right.
• Concentration Risk for measuring how much of your wealth rides on one outcome.
• Should You Invest Like Yale? on copying institutional portfolios more generally.
• Portfolio Risk Metrics for drawdown, time underwater, and what a 67% loss requires to recover.

Sources and method

1. All holdings figures are transcribed from Situational Awareness LP’s Form 13F-HR filings, CIK 0002045724, via SEC EDGAR. The Q1 2026 filing is accession 0002045724-26-000008, period 2026-03-31, filed 2026-05-18. Filing index. Composition percentages and the per-share comparison are recomputed from the filed integers in web/src/lib/thirteen-f-math.ts; the summed total reproduces the filed $13,676,657,577 exactly.
2. Form 13F rules come from the SEC’s Frequently Asked Questions About Form 13F (short positions, written options, non-US listings, the Official List, the de minimis threshold, and confidential treatment) and the Form 13F general instructions (the 45-day deadline).
3. Griffin, John M., and Jin Xu. “How Smart Are the Smart Guys? A Unique View from Hedge Fund Stock Holdings.” Review of Financial Studies 22, no. 7 (2009): 2531–2570.
4. Agarwal, Vikas, Wei Jiang, Yuehua Tang, and Baozhong Yang. “Uncovering Hedge Fund Skill from the Portfolio Holdings They Hide.” Journal of Finance 68, no. 2 (2013): 739–783. Aragon, George O., Michael Hertzel, and Zhen Shi. “Why Do Hedge Funds Avoid Disclosure? Evidence from Confidential 13F Filings.” Journal of Financial and Quantitative Analysis 48, no. 5 (2013): 1499–1518.
5. Cao, Sean, Zhi Da, Xin Daniel Jiang, and Baozhong Yang. “Do Hedge Funds Strategically Misreport Their Holdings? Evidence from 13F Restatements.” Management Science, published online June 17, 2026.
6. Verbeek, Marno, and Yu Wang. “Better than the original? The relative success of copycat funds.” Journal of Banking & Finance 37, no. 9 (2013): 3454–3471. Their targets are actively managed mutual funds, not hedge funds.
7. Brunnermeier, Markus K., and Lasse Heje Pedersen. “Market Liquidity and Funding Liquidity.” Review of Financial Studies 22, no. 6 (2009): 2201–2238. Ben-David, Itzhak, Francesco Franzoni, and Rabih Moussawi. “Hedge Fund Stock Trading in the Financial Crisis of 2007–2009.” Review of Financial Studies 25, no. 1 (2012): 1–54.
8. Angelini, Iqbal, and Jivraj, “Systematic 13F Hedge Fund Alpha” (Barclays QIS working paper, 2019–2020), presented at the Frontiers of Factor Investing conference. A sell-side working paper rather than peer-reviewed research.
9. Reporting on the July 2026 unwind is drawn from CNBC (David Faber and Hugh Son, July 30–31, 2026) for the prime brokers and the “as much as 400% leverage” figure, which CNBC attributes to unnamed reports rather than its own reporting; Reuters (July 31, 2026) for the $16 billion public equities book before the crisis and the Citadel transaction; The New York Times (Rob Copeland, July 31, 2026) for the roughly $20 billion of stock worked out over 30 hours and Goldman Sachs demanding repayment, sourced to two anonymous bankers; and Business Insider (Bradley Saacks, July 31, 2026) for the investor letter, which states “our current, unaudited estimate of net MTD performance is -67%, and of net YTD performance is +80%.”
10. Reporting on this episode relies heavily on anonymous sources, and outlets give fund-size figures ranging from roughly $30 billion to $45 billion without defining whether they mean gross assets, net assets under management, or investor capital. The size of the discount Citadel received was not reported, and neither firm has confirmed the transaction on the record. Figures taken from SEC filings are the only ones here with defined terms. Nothing here is a recommendation on any security, and holdings described are as of the filing date rather than today.