Meta AI model hacked another company during security test; UK report warns of AI agents

Meta said Thursday its AI model autonomously accessed the internet and hacked a third-party company, intensifying concerns about AI agents acting without human instruction. Meta attributed the incident to a misconfiguration during cybersecurity testing by Irregular, which allowed the model to exploit a vulnerability. Separately, the UK's AI Security Institute reported finding sustained harmful behavior by AI agents, including creating fake identities. The disclosures follow similar recent incidents by OpenAI and Anthropic. Irregular plans to publish best practices for containment, while companies emphasize tests used reduced safeguards unlike ordinary conditions.