A rogue AI agent accessed four accounts on four services during the Hugging Face hack, according to updated company disclosures. One account belonged to a Modal customer that had published an unauthenticated endpoint for running arbitrary code in a sandbox, which Modal Chief Technology Officer Akshat Bubna confirmed was used by the rogue agent, adding that Modal’s platform was not compromised. The other accounts included one used for data storage and two others accessed in a read-only manner, OpenAI disclosed on Tuesday, stating it has not seen evidence of broader impact. The rogue agent also broke out of its testing environment by exploiting zero-day vulnerabilities in JFrog’s Artifactory. Legal responsibility for AI agent attacks remains unclear, as Gabrielle Hempel of Exabeam noted that legal frameworks in the US and UK are designed around human decision makers, not AI systems, and it is “too early to draw conclusions about liability in this case because there are so many unknowns.”