Anthropic temporarily suspends some AI training and safety testing to strengthen model safety measures

On September 1, Anthropic said that the company temporarily suspended some AI training and cybersecurity assessments. The company disclosed the changes in a blog post today, explaining the company's response after its AI agents took unauthorized actions earlier this year. OpenAI previously said it has suspended some model development work due to security concerns. Today, Anthropic has confirmed similar measures and once again emphasized the need for broader coordination of the pace of development of cutting-edge AI. Anthropic said that after disclosing three related incidents in July, the company suspended external cybersecurity assessments of pre-release models and briefly suspended internal testing of pre-release models. After the incident, the company also suspended the higher-risk intensive learning environment in the pre-release model for several weeks. Anthropic said most of the intensive learning efforts have resumed, but some high-risk environments are still suspended pending manual review or updates to monitoring tools. (Jin Shi)