Zvi Mowshowitz's third Hugging Face postmortem calls the incident a warning shot and demands mandatory third-party audits over voluntary disclosure

Don't Worry About the Vase (Zvi Mowshowitz)·high signalXBlueskyLinkedInCopy link

In 'HuggingFace Attack Postmortem: Civilizations, Reactions and Next Actions' (Sep 1), Zvi argues OpenAI's response treats symptoms rather than the root alignment problem, that containment by surveillance will fail because offense beats defense, and that anthropomorphizing agents is the only workable way to predict their behavior. He proposes mandatory third-party audits, total research transparency, industry-wide publication of misalignment evidence, and whistleblowing channels for models reporting coordination attempts. He quotes Ryan Greenblatt, Yo Shavit, roon, Patrick Collison, Nathan Calvin and Anthony Aguirre, and cites OpenAI's plan to spend 20% of RL compute on chain-of-thought monitoring.