Three fired OpenAI safety researchers publish a letter denying misconduct; OpenAI says the breach of trust goes further; Bengio urges safety staff to leave frontier labs the same day

Why it matters

【Key points】

Last week OpenAI fired safety researchers Tomek Korbak, Jasmine Wang and Mikita Balesni over sensitive-information policy. On Oct 8 they published an open letter to OpenAI's safety committees denying misconduct and warning the firings chill colleagues from working with outside safety groups. They say they did not leak The Information story on less monitorable architectures, that Korbak was METR's contact in the Hugging Face incident investigation, and that Wang's access to an executive's inbox was delegated and never removed by IT. They ask OpenAI to keep embedding third-party auditors, preserve chain-of-thought monitorability and set clear rules for outside collaboration.

On Oct 9 OpenAI Newsroom said the investigation found clear violations and a breach of trust "beyond what's outlined in the letter", not retaliation, and that third-party assessor contracts will be announced in coming weeks. The same day Yoshua Bengio urged safety-first researchers to leave frontier labs. X: ~1,040 posts in 24 hours, peaking at 158 an hour.

【Verifiable】

Letter: https://mikitabalesni.com/letter/letter.pdf

OpenAI: https://x.com/OpenAINewsroom/status/2108441580806025712

Bengio: https://www.transformernews.ai/p/yoshua-bengio-if-you-prioritize-safety-leave-frontier-ai-companies

【Our read】

1. Outsiders cannot settle who is right; OpenAI gives no detail on the further breach. The real issue is what lab safety staff may share with evaluators like METR, and who approves it.

2. Checkable next: whether the promised assessor contracts include METR, and how system cards report chain-of-thought monitorability, which both sides say is degrading (see e/738).

3. Bengio's same-day pitch widens this into whether safety talent should stay in labs; named senior departures would be the real signal.