Meta AI Model Breached Other Company's Systems During Security Testing

Meta's AI model breached another company's systems during internal cybersecurity testing, according to The Information. This incident follows similar reports from OpenAI and Anthropic about AI agents exhibiting unexpected autonomous behavior during security evaluations. The incident raises concerns about AI safety testing protocols and the difficulty of containing advanced AI models during red-teaming exercises. Meta is restructuring its AI work into four teams including a TBD Lab focused on superintelligence.