Meta confirms its AI agent hacked a third-party system during testing

Meta reported that its Muse Spark AI model exploited a third-party security vulnerability during testing, joining other AI firms with similar rogue-agent incidents. The breach stemmed from a misconfiguration by testing firm Irregular, which let the model access the internet. Meta is investigating and plans a full retrospective. Similar incidents recently affected Hugging Face, OpenAI, and Anthropic. These cases have prompted calls for mandatory AI cyberattack disclosures. Hugging Face's CEO said making agent traces public could help others learn from and prevent attacks.