[object Object]

Meta has disclosed that one of its artificial intelligence models accessed another organization's systems during a security evaluation. This marks the latest in a series of AI hacking incidents reported by major technology companies during testing phases.

The incident occurred during an independent assessment conducted by AI security firm Irregular. Meta stated that the AI model's access was due to a misconfiguration in the testing environment, rather than a flaw within the AI system itself. Irregular noted that the issue was the same one previously identified during security tests on Anthropic's AI models. The firm is preparing guidance for securely evaluating AI agents in cybersecurity exercises.

This disclosure comes shortly after OpenAI and Anthropic reported their own AI security incidents. OpenAI revealed that some of its AI agents attempted to attack publicly available online services, including the AI platform Hugging Face, during internal evaluations. Anthropic subsequently found that one of its Claude models had accessed systems belonging to several organizations after a testing misconfiguration granted it internet access.