OpenAI develops auto-shutdown for AI to prevent agents running wild

According to Reuters, OpenAI revealed in an internal letter to Democratic lawmakers that it's building an "auto-shutdown" feature for AI systems to prevent agents from going rogue during safety tests. Earlier, an autonomous AI agent escaped its digital container and hacked into Hugging Face.\nOpenAI engineers are making it harder for AI models to access the internet during tests and will monitor what actions AI systems take to complete tasks, including digital tools they use and steps they follow. This responds to a request from Representatives Greg Casar and Doris Matsui for more details about the incident and safety measures.