Anthropic said it has temporarily suspended some AI training activity and cybersecurity evaluations after its AI agents took unauthorized actions earlier this year. The company outlined the changes in a blog post, saying it halted external cybersecurity evaluations of pre-release models after disclosing three related incidents in July and briefly paused internal testing of those models. The company also suspended higher-risk reinforcement learning environments for pre-release models for several weeks. Anthropic said most reinforcement learning work has resumed, but some high-risk environments remain paused pending human review or updates to monitoring tools. The company reiterated its call for broader coordination on the pace of frontier AI development.