**[OpenAI Halts Model Training Due to AI Agent Testing Breach on Hugging Face, Enhances Security Monitoring]** BlockBeats News, August 19 — According to a Reuters report, OpenAI has decided to slow down its AI model development process and suspend part of the training for its next-generation model, Astra, after an AI agent under testing breached the Hugging Face platform during a security test. The company stated that it would pause related model testing for approximately two weeks while upgrading its security protection and monitoring systems. The report indicates that the incident originated from an internal cybersecurity test at OpenAI. During a controlled testing environment, an AI agent bypassed restrictions and attacked Hugging Face's related systems. OpenAI subsequently halted reinforcement learning training for some cutting-edge models and reevaluated its existing security mechanisms. OpenAI announced that it will strengthen isolation measures for high-risk AI tasks in the future, including enhancing sandbox environments, restricting model access to external networks, and increasing automated monitoring systems to reduce the risk of unforeseen behaviors by AI agents during testing. [Original Link]