← Back to Model Beat
Research·6d ago·all news from November 16, 2026

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

Related stories

ResearchResearch acceleration: The jagged frontierNov 17 · 4 sourcesResearchTop AI Models Regularly Identify Civilians as Military Targets in TestsNov 17 · 4 sourcesResearchDepth Pro 2: Sharp and Temporally Consistent Metric Monocular DepthNov 18 · 5 sourcesResearchGoogle, Salesforce, Sarvam AI partner with Karnataka to train 63,200 people in AINov 17 · 12 sources