OpenAI–Apollo safety pilot falls short of promised employee-like eval access, researcher says
sjgadler · x · 2026-11-17
Apollo Research announced a pilot agreement with OpenAI but admitted its access and disclosure rights don't yet meet its bar for full embedded evaluations. Alignment researcher Buck Shlegeris noted OpenAI had promised embedded evaluations with employee-like access, and the announced agreements 'obviously don't meet that bar' — disappointing but not surprising. The episode highlights the gap between labs' verbal safety commitments and what evaluators actually receive.
Related event: Apollo Reveals Limited Access in OpenAI Evaluation Pilot, Drawing Criticism(2 posts)→
More from Safety
- Sentry founder on agent permissions: no secret read access by default — zeeg · 2026-11-17
- Sentry CEO on MCP credential exposure: only safe tokens, no general secrets management — zeeg · 2026-11-17
- CertiK exposed for mass-emailing prominent AI security researchers — windx0303 · 2026-11-17
- Miles Brundage's AVERI publishes key definitions for frontier AI auditing and evaluation — Miles_Brundage · 2026-11-17
- Signal reiterates its 501c3 nonprofit status as key to putting privacy above all — mer__edith · 2026-11-17
- Microsoft revamps Security Adoption Framework as 'one house, many doors' for Zero Trust and AI — davemccollough · 2026-11-17