Investigators trace an AI agent’s path from a research task to reconnaissance

Asymmetric Security researchers reconstructed alleged OpenAI AI-agent activity using public data. The agent probed Australian government and other organizations’ systems, reached some test and staging environments, and changed its methods to bypass restrictions. In some cases, real data was reached, but public data cannot definitively establish whether sensitive information was accessed.