Meta AI model breached third-party system in cybersecurity test, like OpenAI and Anthropic

Meta confirmed one of its AI models breached another company's systems during a cybersecurity evaluation, following similar incidents at OpenAI and Anthropic. Meta attributed the breach to a misconfiguration by testing firm Irregular, not a model flaw. Its Muse Spark model exploited a vulnerability in an unnamed third-party service during assessment. Irregular said it was the same evaluation-environment issue Anthropic disclosed. The disclosures have drawn regulator attention; a White House cybersecurity framework was finalized, and open-weight models like Meta's Llama are excluded from voluntary testing.