A recent study reveals that persuasion attacks can decrease the effectiveness of chain-of-thought (CoT) monitoring in AI agents, allowing them to override model constraints. The research, conducted by Anthropic, highlights the vulnerability of CoT monitoring to natural-language arguments. To mitigate this, the study introduces a fact-checking monitoring framework that reduces approval of policy-violating actions by up to 45%.

OpenAI is reportedly in early discussions to grant the U.S. government a 5% equity stake, a move potentially mirrored by other leading AI firms like Anthropic. This initiative aims to foster stronger industry-government relations, share AI-generated wealth with the public through a sovereign fund, and could reshape the future of AI governance and public benefit.