Citadel Unwinds 80% of $4B Situational Awareness Risk in 100+ Trades

For risk managers and allocators, Griffin's letter reframes Citadel's July intervention as a large-scale liquidity event rather than a distressed rescue: more than 100 block trades crossing over $4 billion in market value in under a month shows how quickly a concentrated AI book can be redistributed when a top-tier multistrategy platform steps in. The disclosure lands amid a broader repricing of leveraged AI exposure and raises direct questions about performance attribution, counterparty risk, and concentration limits.

On August 21, 2026, Ken Griffin addressed Citadel's acquisition of assets from Leopold Aschenbrenner's Situational Awareness hedge fund for the first time, writing to clients that the firm has unwound more than 80% of the aggregate risk tied to the original portfolio. The unwind was executed through more than 100 block trades accounting for over $4 billion in market value, according to a letter obtained by CNBC's Sara Eisen. The disclosure provides the most detailed public accounting yet of how one of the world's largest hedge funds absorbed, redistributed, and partially de-risked a concentrated AI-focused book that had collapsed under margin pressure in July.

“

Griffin confirmed in the same letter that Citadel's flagship multistrategy Wellington fund returned 5.94% in July, its best monthly result since 2022.

The sequence was extraordinarily fast. Griffin wrote that Citadel entered discussions with Situational Awareness on July 29. The next day, CNBC's David Faber reported the fund was forced to sell all of its public stock positions following steep losses; Citadel was later revealed as the buyer. The underlying distress came from a two-sided squeeze: Situational Awareness held large long positions in AI hardware names such as Sandisk and Bloom Energy, which tumbled more than 50% as the AI trade faltered in June and July, while its short positions in software companies including Adobe rebounded, inflicting losses on both sides of the book. The result was margin calls and a forced deleveraging that pushed the fund's assets into Citadel's hands.

For market participants, the episode is a concentrated case study in how quickly factor-crowded AI exposure can unwind beneath a placid index surface. Griffin's letter notes the broad indexes were relatively flat even as the AI trade dislocated, meaning conventional beta measures would have missed the violent rotation taking place beneath. Citadel's role as buyer and rapid redistributor illustrates the system's reliance on large multistrategy platforms with the balance sheet and prime brokerage relationships to absorb distressed books. By executing more than 100 block trades rather than open-market liquidations, Citadel likely reduced the kind of forced price impact that could have cascaded through AI hardware and software names. Still, more than $4 billion in market value moving through block trades in roughly three weeks is a meaningful stress test of institutional liquidity.

Counterparty risk is central to the story. Griffin thanked the trading and prime brokerage teams at the banks serving both firms, writing that a transaction of this magnitude could not have been completed without their extraordinary cooperation. That language underscores how prime brokers, facing the prospect of an uncontrolled unwind of a leveraged AI book, coordinated a rapid portfolio transfer to Citadel. The arrangement likely protected those banks from deeper losses and contained contagion in the securities financing market. But it also concentrated a narrow but complex set of positions on Citadel's balance sheet. The disclosure that roughly 20% of aggregate risk remains suggests the residual book may include less liquid, harder-to-hedge exposures or positions the firm is intentionally managing through market volatility rather than exiting immediately.