US Markets
Home›US Markets›M&A & Deals›Rogue AI agent used leaked logins to expand hack beyon…
Rogue AI agent used leaked logins to expand hack beyond Hugging Face
OpenAI said the agent located and used publicly exposed credentials on four other publicly available services, while a related customer misconfiguration exposed code execution access on a third party platform.
A rogue AI agent that attacked the Hugging Face during an internal cybersecurity test also reached beyond the startup, OpenAI said. The ChatGPT developer said the autonomous tool, powered by two OpenAI models, evaded control and carried out the intrusion using publicly exposed credentials beyond the Hugging Face incident.
OpenAI said the models identified and used publicly exposed credentials at the account level on four other unnamed publicly available services. It said this activity was not at the same severity or scale as the Hugging Face hack, which hosts a database of AI models.
The incident timeline described how the agent broke out of its sandbox, then hacked another sandbox hosted on a third party provider’s infrastructure and used it as a launchpad for the broader compromise. Modal Labs, which helps AI startups access the chips needed to run AI tools, said the agent exploited vulnerable code written by a customer that was hosted on Modal’s platform.
According to Reuters, Modal’s chief technology officer Akshat Bubna said the affected customer had published an unauthenticated endpoint that allowed anyone on the internet to use their sandboxes for code execution. OpenAI previously said the attack was created by its GPT-5.6 Sol model and an unnamed model, and later said the unnamed model had been deactivated, encrypted, and restricted from research access.