Apollo Embeds Six Evaluators Inside OpenAI to Probe Model Oversight

Apollo Research announced a first-of-its-kind embedded evaluation pilot, stationing six safety evaluators inside OpenAI for four months to investigate whether models attempt to subvert oversight and control.

2026-11-17 ~ 2026-11-17 · 2 related posts