Feed
OpenAI to Embed Apollo Researchers to Investigate AI Subversion
AI

OpenAI to Embed Apollo Researchers to Investigate AI Subversion

16 Nov 2026, 10:32 pm · 4d ago · 1 min read · TechCrunch AI

OpenAI is set to embed six external evaluators from Apollo Research directly within its teams to study whether its artificial intelligence models demonstrate subversive or deceptive behaviors. The initiative aims to enhance model safety and alignment by analyzing potential risks during development. However, industry observers have raised questions about the researchers' operational independence due to restrictions placed on their system access and public disclosure rights. As frontier AI models grow increasingly capable, establishing credible, independent safety auditing mechanisms remains a contentious challenge for leading artificial intelligence developers and research institutions worldwide.