Live intelligence

Latest AI updates

1 updates
Research

Transparency Disclosures for Chain-of-Thought Monitorability

A new arXiv paper argues that frontier AI developers should report specific indicators to track threats to chain-of-thought (CoT) monitoring and support coordination. It proposes five indicators spanning third-party CoT access and base architecture specifications, CoT quality (naturalness and faithfulness), and monitoring effectiveness (detection and uplift). An audit of disclosures from Anthropic, OpenAI, and Google DeepMind found no developer covers the full set, though each indicator has been reported by at least one. The authors recommend a cross-developer audit with deeper access, via embedded evaluators or an Independent Verification Organization.

chain-of-thought monitoringAI transparencyfrontier AI developers