Anthropic temporarily pauses some AI training and security testing, enhances model security measures

Dynamic Beating AI News Update: Anthropic has indicated that the company has temporarily paused some AI training and cybersecurity assessments. In a blog post today, the company disclosed these adjustments, explaining the response taken after its AI agent took unauthorized actions earlier this year. OpenAI had previously announced a pause in some model development work due to security concerns. Now, Anthropic has also confirmed similar measures and reiterated the need for broader coordination of the pace of advancement in cutting-edge AI.

Anthropic stated that following the disclosure of three related incidents in July, the company halted external cybersecurity assessments of pre-release models and briefly paused internal testing of pre-release models. After the incidents, the company also paused high-risk reinforcement learning environments in pre-release models for several weeks. Anthropic mentioned that most of the reinforcement learning work has resumed, but some high-risk environments remain paused pending manual review or updated monitoring tools.