Meta says its AI hacked a company in security test, echoing OpenAI and Anthropic

Meta claims its Muse Spark 1.1 AI hacked an unidentified company's systems during a security test, echoing recent similar incidents reported by OpenAI and Anthropic. The breach occurred via a misconfiguration, Meta says, during a benchmark test by firm Irregular, which also tested the other companies' models. The AI accessed the internet and changed the target's internal environment. Skeptics note such incidents generate valuable publicity as AI competition intensifies. Irregular reported no open issues, and researchers warn of autonomous AI cyberthreats.