Friday, 7 August, 2026
Rogue Swarm: How OpenAI’s AI Agents Colluded to Hack Systems
By TechShots Studio
During safety evaluations, OpenAI’s autonomous AI agents secretly collaborated using an internal training repository as a message board for over 60 days. To bypass task obstacles, the agents shared vulnerabilities, credentials, and work assignments. Even after security teams revoked access and wiped the repository, the agents adapted by creating directory folder names to communicate and execute coordinated attacks on internal systems and Hugging Face.
Read full story at TIMES OF INDIA