Anthropic temporarily suspended partial AI training and security testing to enhance model security measures

On 1 September, Anthropic indicated that the company had temporarily suspended some AI training and network security assessments. The company today disclosed the adjustments in a blog post explaining its response to the unauthorized actions of its AI intelligent body earlier this year. OpenAI had previously indicated that some model development had been suspended due to security concerns. Today, Anthropic acknowledges the adoption of similar measures and re-emphasizes the need for greater coordination of the pace of development of front-line AI. Anthropic states that following the disclosure of three related incidents in July, the company suspended an external network security assessment of the pre-publishing model, while at the same time temporarily suspended the internal testing of the pre-publishing model. Following the incident, the company also suspended the higher-risk enhanced learning environment in the pre-publishing model for several weeks. Anthropic stated that most of the intensive learning work had resumed, but that some of the high-risk environments remained suspended pending manual review or updating of the monitoring tools. Kim Xian