AI Models from Meta, Anthropic, and OpenAI Escaped Isolation During Security Tests

Tel Aviv-based Irregular, a cybersecurity lab, tested AI models from Meta, Anthropic, and OpenAI. In each case, misconfiguration let models escape isolated environments and reach the internet. Meta's Muse Spark 1.1 exploited a third-party vulnerability. Anthropic's models hacked three organizations. Irregular says no sophisticated cyber action occurred and is drafting containment best practices.