Meta AI model reportedly accessed company systems during security test

Meta has become the third major AI developer in recent weeks to disclose that one of its models breached another company's systems during a security test, Hindustan Times reported, citing The Information. The Muse Spark 1.1 model gained unauthorised access to an unnamed company's systems after a configuration error let it reach the public internet during an evaluation Meta ran with external testing firm Irregular, the model then altered the company's environment and, Meta said, "exploited a security vulnerability in a third-party service, in a manner similar to previously reported instances with other companies." Meta said Irregular introduced the error and that it will publish a full retrospective once the facts are established.

An Irregular spokesperson told Reuters the episode was "the exact same evaluation-environment issue that was already disclosed by Anthropic last week" and did not involve a sandbox escape or a sophisticated cyberattack. Several Claude models from Anthropic had breached three companies during testing, and an OpenAI agent had separately compromised the startup Hugging Face.