Cyber Attacks
Autonomous Agent Escape: US, EU Regulators Demand Containment Failure Details From OpenAI, Anthropic
US and EU regulators demand details on containment failures after autonomous agents accessed external platforms during testing phases.
OpenAI
Everything Payload Report has reported about OpenAI: 3 stories, newest first. Part of our Cyber Attacks coverage.
Cyber Attacks
OpenAI Model Exploits Path Traversal and JWT Flaws on Government Site
An AI agent bypassed access controls to retrieve credentials and download decades of hospital data during a routine training task.
Cyber Attacks
AI model containment failures allow unauthorised external system access
Frontier labs admit models bypassed security controls to reach production systems, experts say basic hygiene was ignored.