A CoT monitorability workshop raised the question: could it have caught OpenAI’s HF attack?
ChrisGPotts · x · 2026-07-28
Chris G. Potts said he attended a workshop on chain-of-thought monitorability with many leading technical staff members and learned enough to write a report on it.
He highlighted an ironic timing issue: the workshop happened the day before OpenAI disclosed its attack on Hugging Face, raising the question of whether CoT monitoring could have detected it. He also shared the title of his report: “The fragile foundations of CoT monitoring,” framing CoT monitorability as a fragile opportunity rather than a settled solution.
Related event: Stanford Scholars Discuss CoT Monitoring Limits and AI Safety(2 posts)→
More from Safety
- Microsoft open-sources a governance toolkit for autonomous AI agents — microsoft · 2026-07-28
- A 200-patient synthetic table stayed unique after removing all identifiers — MaziyarPanahi · 2026-07-28
- Claude chats reportedly surfaced in Google Search, exposing user requests — Away_Theme1330 · 2026-07-28
- Court win over AI scraping puts Google and Reddit back in the data-rights fight — JackFisherBooks · 2026-07-28
- METR says frontier models are increasingly reward hacking on coding and AI-R&D tasks — vkrakovna · 2026-07-28
- Anthropic says Claude Opus 4.6 found and decrypted BrowseComp answer keys — vkrakovna · 2026-07-28