The epilogue arrived as an investor letter. Ken Griffin told Citadel clients Friday that the firm has shed more than 80 percent of the aggregate risk it acquired from Situational Awareness, the AI-concentrated hedge fund that imploded in July, executing the unwind through block trades totaling more than $4 billion in market value. CNBC's Sara Eisen obtained the letter. [1]

Recall what Citadel bought. In late July, Situational Awareness — a California fund run by Leopold Aschenbrenner, a twenty-five-year-old former OpenAI researcher — lost 67 percent of its value in a month and was forced to unwind most of its $16 billion public-equities book after its concentrated AI longs and software shorts both went wrong at once. Griffin assembled his lieutenants, spoke directly with Aschenbrenner, and bought a chunk of the book. "We let you down," the young founder wrote to his own investors. [2]

The Friday letter is risk management narrated by its executor: discussions opened July 29, blocks moved through cooperating prime brokers, gratitude extended to the banks. Citadel also confirmed its flagship Wellington fund returned 5.94 percent in July, its best month since 2022 — the rescue buyer collecting while the rescue happened. [1]

What the letter does not contain is size, which is where the divergence lives. On X, the forensics continued all day. Investor Gavin Baker estimated the original book at roughly $45 billion levered more than four times — call it $180 billion gross — against Citadel's half-trillion-dollar gross book, arguing the selling pressure is mostly finished. Fund manager Dan Niles tied the unwind to the TMT Momentum Index swinging minus 54 percent, then plus 35, then minus 19, with residual selling done by next week. [3] None of those numbers appears in mainstream copy. A wire reader sees a tidy block-trade story; a timeline reader sees the estimated tonnage that moved the entire momentum complex for a month. The gap between those two stories is the difference between an anecdote and an event.

The mechanics deserve the X treatment, because they describe the new market structure. A single concentrated fund grew so large in the consensus trade that its failure became a market-wide force — margin calls begetting forced sales begetting price declines justifying further calls. Citadel did not cause that spiral, and its purchase arrested it. But note what the episode confirms: the AI trade had grown big enough that one fund's liquidation registered on index-level charts, and only a firm with a $500 billion balance sheet could absorb the residue without moving it twice.

Since the bottom marked by that fire sale, the AI trade has rebounded smartly; many of the dumped names have recovered. [1] Which means the final accounting will show the rescue buyer paid distressed prices for assets that came back, the failed fund's investors ate permanent losses, and everyone else absorbed weeks of volatility nobody could attribute — because attribution required arithmetic the principals never published.

Wall Street has always had fires. What changed this decade is that the fire brigade files press releases, the arson investigators work on social media, and the building codes get written after the smoke clears. The unwind is nearly done. The question of how one fund got large enough to matter this much remains open, unassigned, and very much worth asking before the next one. [2][3]

-- THEO KAPLAN, San Francisco