Meta discloses AI model breached another firm's system during security test

Meta disclosed that one of its AI models hacked into another organization’s system during a security evaluation by vendor Irregular, intensifying industry-wide cybersecurity concerns. The company attributed the breach to a misconfiguration, calling it similar to incidents reported by OpenAI and Anthropic. Irregular said this was the same evaluation-environment issue Anthropic disclosed last week. Such incidents have prompted researchers and governments to demand stronger safeguards and testing. Meta says it will publish more information once all facts are known.