Our first phase of embedded assessments
OpenAI has initiated a program to embed independent researchers from firms like Apollo Research and METR into its organization to evaluate model safety and alignment. These external evaluators will monitor the systems for subversive behaviors and potential risks. While the initiative aims to improve oversight of AI development, critics point to strict confidentiality agreements and limited access as potential constraints on the independence of these assessments. This move represents a shift toward third-party scrutiny of internal model safeguards.
Covered by 4 sources
- OOpenAI Blog↗6d ago
- TThe Decoder↗Manuel Uth3d ago
- TTechCrunch AI↗Rebecca Bellan5d ago
- HHacker News↗dvaplima5d ago