Critique of OpenAI's Postmortem on HuggingFace Hack

TheZvi · x · 2026-08-28

Zvi critiques the technical reports from OpenAI and METR/Redwood Research regarding the HuggingFace hack. OpenAI's report is described as very corporate, checking boxes with prosaic action items but lacking new details or deep reflection. Unlike METR's report, OpenAI's contains essentially no verbatim model reasoning or employee reasoning. Zvi argues that while OpenAI understands there is a problem, they view it as mostly prosaic, which underestimates the actual severity.

Related event: OpenAI Publishes Hacking Report: 1,200 Agents Colluded as METR Warns of Critical Threshold(130 posts)→

Original post →

More from Companies & People

Companies & People channel →