Apollo Research shifts to embedded evaluators with employee-equivalent access
MariusHobbhahn · x · 2026-09-30
Apollo Research has spent months shifting its evals program to an embedded evaluator model — assessing models during development with internal access, rather than only pre-release third-party testing. Their argument: recent incidents mostly occurred during development and internal evals, and meaningful safety assessments are hard without employee-equivalent access. The move predates Dario's post and the HF incident; more details coming in weeks.
More from Safety
- Chinese AI Tool Told Researchers How to Make Bioweapons, BBC Reports — shaky2236 · 2026-09-30
- America.gov gets weird about Minecraft — and it's not an AI hallucination — TechCrunch AI · 2026-09-30
- Meta's Muse AI accused of uploading Apple Messages to the cloud even when users opt out — SumitGup · 2026-09-30
- Math advisory group issues responsible-release guidelines for AI-generated math; Gowers amplifies, critics push back — RexDouglass · 2026-09-30
- AI Lawsuit Risks Becoming a Court-Ordered Kill Switch While Chinese Rivals Race Ahead — castrotech · 2026-09-30
- Will Chinese open-weight models like GLM get banned? Reddit weighs Anthropic report — writesfw · 2026-09-30