Ex-OpenAI researcher: labs will never agree on one third-party safety evaluator

suchenzang · x · 2026-09-13

Former OpenAI researcher suchenzang reiterates her view that labs place near-zero probability on agreeing to a single 'third-party safety evaluator', because such a body would need to simultaneously be competent enough to run real frontier evals (not repackage existing ones), financially independent of existing labs, and have the backbone to speak up when things go wrong. She adds the cryptic hint 'keyword: embedded' pointing at related news.

Related event: Ex-OpenAI Researchers: Unified Third-Party AI Safety Evaluators Nearly Impossible(3 posts)→

Original post →

More from Models

Models channel →