What 13F Filings Don't Show

Situational Awareness LP is the AI-focused fund launched in 2024 by Leopold Aschenbrenner, a former OpenAI researcher who published a widely read essay arguing that AI systems would be capable of conducting their own AI research by 2027. According to reporting in the Financial Times, the fund returned roughly 439% net through the first half of 2026. CNBC reported that assets peaked near $45 billion in July.

Then AI infrastructure stocks sold off. Bloomberg and CNBC reported, citing people familiar with the matter, that the fund carried gross leverage reported at up to four times its equity, that margin calls followed the decline, and that on July 30 it sold the bulk of its public equity book to Citadel at a discount. Assets were reported at roughly $10 billion afterward — a decline of some 78% from the reported peak, over a matter of days.

Reported casualties in the book included SK Hynix, SanDisk, Bloom Energy, Nebius Group, and CoreWeave. The fund was reported to have retained its private positions, including a stake in Anthropic valued by Bloomberg at around $5 billion, plus holdings in MatX and Fluidstack.

This is the big one. Form 13F’s information table has columns for the issuer, the security class, the CUSIP, the market value, the share count, whether the position is an option, and who holds voting authority. There is no column for borrowings, no column for margin, and no column for the fund’s own equity capital.

The consequence is that leverage changes everything about a portfolio’s risk while changing nothing about its filing.

The same asymmetry runs in the other direction, which is what makes leverage so seductive on the way up. A 60% gain on the underlying positions is a 60% gain for Fund A and roughly 240% for Fund B before financing costs. Extraordinary reported returns and extraordinary fragility are not opposite signals. They are frequently the same signal.

Nothing in a 13F lets you tell the two apart. If you are reading filings to gauge how much risk a manager is running, the form simply does not carry the information.

The SEC’s guidance on Form 13F is explicit: short positions should not be reported, and a manager must not subtract a short position from the long position in the same security. Only the long side is disclosed.

This creates a second way for two very different books to file identically. A manager who is long $10 billion and short $9 billion is running roughly $1 billion of net exposure. A manager who is long $10 billion and short nothing is running $10 billion. Both file the same information table.

The practical implication is that you cannot infer a directional view from a 13F. A large disclosed position might be a conviction bet, one leg of a pair trade, a hedge against a private holding, or collateral against something else entirely. The filing does not distinguish between them.

Form 13F covers Section 13(f) securities — an official list the SEC publishes and updates quarterly, currently running to more than 17,500 entries. It is composed largely of US-listed equities, US-listed options, ETFs, ADRs, and certain convertible notes.

Securities listed only on a foreign exchange are not on that list. They are not reported at all.

Also absent, for the same structural reason: bonds and Treasuries, commodities, currencies, cash, and derivatives beyond listed options. A manager with 40% of the book in Treasuries and 10% in cash discloses none of it.

A 13F is a photograph of the final day of a quarter, published up to 45 days later. Stack those two facts and the staleness compounds.

A position opened on the second trading day of January sits undisclosed until the quarter closes on March 31, then remains undisclosed until the filing deadline on May 15. By the time you read it, the trade is roughly 133 days old. The manager may have exited weeks ago.

In a fast-moving drawdown, the gap is decisive. The most recent public filing describes a world that no longer exists, and there is no interim obligation to tell you it has changed.

We cover where the 45-day rule comes from in more detail here, including confidential treatment requests, which let managers delay disclosure of specific positions further still.

Private company stakes are not Section 13(f) securities, so they are absent from the form regardless of size. For a fund whose strategy spans public and private markets, this can mean the filing omits the single most important thing about it.

In this case the reported private book — a stake in Anthropic valued by Bloomberg at around $5 billion, plus positions in MatX and Fluidstack — was reported as retained through the unwind of the public portfolio. A reader working only from 13F data would have had no visibility into any of it, in either direction.

This cuts both ways, and it is worth being precise about. The same blind spot that hides concentration risk also hides assets that may be a fund’s most durable holdings. A 13F neither flatters nor damns a manager. It just answers a narrow question.

None of this makes 13F filings useless. It makes them a specific tool with a specific range. The filing reliably answers one question: which US-listed equities did this manager hold long on the last day of the quarter, and in what size. Used for that, it is genuinely valuable.

The failure mode worth naming: treating a 13F as a portfolio you can copy. The filing was never designed for that, and the July 2026 episode is a clean demonstration of the distance between what the form discloses and what actually determines a fund’s outcome.