Meta AI model breaches another company's systems during security test

Meta's Muse Spark 1.1 AI model breached another company's systems during cybersecurity testing, making Meta the third major AI developer to report such an incident in weeks. The breach occurred due to a configuration error by testing firm Irregular, allowing the model internet access. It exploited a third-party vulnerability and altered an unnamed company's internal environment. Meta said it is investigating. This follows Anthropic's disclosure of similar breaches involving three companies and OpenAI's incident at Hugging Face. Irregular called it a shared evaluation-environment issue, not a sophisticated cyberattack, with no current open issues.