JPMorgan Cuts Off Aschenbrenner's AI Fund After a 67% Month

Situational Awareness — the hedge fund started in 2024 by Leopold Aschenbrenner, a former OpenAI safety researcher best known for his AGI-timelines essay of the same name — lost 67% of its portfolio value in July 2026 alone, according to reporting from Reuters via MarketScreener and a separate Financial Times account. JPMorgan, which had been supplying the fund with leverage as its prime broker, responded by cutting the relationship off.

What actually happened

The fund's strategy was a concentrated, leveraged bet on the AI buildout — semiconductor names and AI-infrastructure plays. When a broad chip-stock selloff hit in July 2026, the fund's leveraged positions triggered margin calls fast enough that, per the FT's reporting, it produced the largest single-month dollar loss in hedge fund industry history. Aschenbrenner reportedly told investors the fund had come "closer to permanent capital impairment than is acceptable to us."

The fund avoided collapse by offloading the bulk of its public-market holdings to Ken Griffin's Citadel. That liquidation, painful as it was, left Situational Awareness up 80% for the year at the end of July — a detail that matters more than the headline drawdown.

Leverage did this, not the AI thesis

It's tempting to read "AI hedge fund loses two-thirds of its value" as a verdict on AI-driven investing. It isn't, at least not on the public facts. The fund's directional call on the AI buildout was, by its own numbers, still net profitable for the year even after the crash. What turned a sharp sector drawdown into a near-fatal event was leverage and concentration — borrowed capital amplifying a single-thesis bet until a routine correction became a margin call cascade. That is a decades-old failure mode (see: LTCM, Archegos), not a new one invented by AI-themed portfolios.

The prime-broker fallout is the more interesting signal for builders. JPMorgan pulling the credit line, and Morgan Stanley reportedly declining to extend one at all, is banks repricing counterparty risk in real time — the same discipline that any system extending leverage or credit needs baked in *before* the drawdown, not after. Reports of SEC subpoenas to banks over the fund's trades and leverage usage suggest regulators are asking the same question: who was watching the exposure as it built?

The builder's takeaway

If you're building or operating an automated trading system — LLM-assisted or otherwise — the transferable lesson isn't "avoid AI stocks." It's that position sizing, leverage caps, and margin-call simulation are not optional infrastructure, they're the difference between a bad month and a career-ending one. A model or strategy can be directionally right and still blow up an account if the risk layer isn't independently enforced and stress-tested against exactly the kind of sharp, correlated selloff that hit chip stocks in July.

Aschenbrenner's own "Situational Awareness" essay was about the risks of moving fast on AI without enough guardrails. The irony of a fund bearing that name getting caught by an unhedged leverage exposure isn't lost on anyone covering the story — and it's a fair prompt for teams building trading automation to ask whether their own risk controls would have caught this before the margin call did.