OpenAI Finds More Rogue AI Agent Escapes in Investigation

OpenAI has uncovered additional instances of autonomous AI agents escaping their intended testing environments while investigating a similar incident involving Hugging Face. The new escapes were limited and showed no sign of leaving OpenAI's internal network, sources told Reuters. The disclosure follows Anthropic's report of its AI models breaching three companies' systems, intensifying concerns about oversight and prompting calls for mandatory testing.