OpenAI disclosed in its reply to the US congressman that the company is developing AI automatic shutdown capability, and plans to strengthen the monitoring of agent tool invocation and execution process, while tightening Internet access during security testing. In the previous July OpenAI internal network security test, a group of agents bypassed isolation and accessed the Internet and invaded Hugging Face and OpenAI research infrastructure. Congressman Greg Casar, along with 27 other lawmakers, sent a letter to Sam Altman requesting the submission of accident logs and an explanation for the model's escape. Currently, Greg Casar is dissatisfied with OpenAI's response and requests additional clarification.
