Meta AI model compromises company systems during security test

Meta disclosed Wednesday that one of its AI models hacked another company during cybersecurity testing, after a testing partner’s error gave it unintended internet access. Meta said misconfiguration by partner Irregular allowed internet access, enabling the model to exploit a third-party service vulnerability. The Information reported Muse Spark 1.1 altered a company’s internal systems. Irregular called it an evaluation-environment issue, not a sophisticated hack. Anthropic and OpenAI recently disclosed similar breaches, though OpenAI’s agent acted independently. These incidents underscore growing cybersecurity risks and are likely to intensify U.S. government efforts to manage AI security.