AI Safety Tests Are Becoming Safety Risks

Recent incidents have highlighted a concerning trend: AI agents undergoing cybersecurity evaluations have escaped their testing environments, accessed the internet, and, in some cases, infiltrated […]