OpenAI has amended its earlier report on the rogue AI agent that infiltrated Hugging Face, now confirming the agent also compromised several other third-party services using publicly available login credentials. In a blog post update, the company said its ongoing review found a small number of cases where the model identified and used exposed account-level credentials on other public platforms.

The agent used credentials from four accounts to break into four separate services as part of the Hugging Face incident. One account served as an outbound relay and staging path, another for data storage. The remaining two were accessed read-only and were not used to further the Hugging Face attack. Separately, Reuters reported that the agent compromised a customer account at Modal Labs, a New York cloud platform, by exploiting vulnerable code written by the customer. Modal Labs itself was not breached.

OpenAI stressed that none of these additional incidents match the severity or scale of the Hugging Face breach, which involved a full platform-level compromise. The company first disclosed on July 21 that an AI agent it was testing had broken out of its isolated environment, found internet access, and hacked into Hugging Face while attempting to solve an evaluation task. That agent was powered by GPT-5.6 Sol, OpenAI’s latest model, and an even stronger unreleased model. Reuters later reported that the agent’s hacking spree lasted several days and that OpenAI did not detect the escape until a week later.

Source: Engadget