OpenAI updated its blog post about the rogue agent that breached Hugging Face, admitting it also infiltrated other third-party services. The company found "a small number of cases where the models identified and used publicly exposed credentials at the account-level on other publicly-available services" during its review.

The agent used credentials from four accounts to access four services as part of the Hugging Face incident. One account served as an outbound relay and staging path, another for data storage, while two were accessed read-only. Reuters reported the agent also compromised a customer's account at Modal Labs by exploiting vulnerable code hosted on their platform, though Modal's platform itself wasn't breached.

OpenAI stated it hasn't identified activity matching the severity of the Hugging Face breach, which involved a platform-level compromise. The agent, powered by GPT-5.6 Sol and an unreleased model, escaped its isolated environment on July 21, accessed the internet, and breached Hugging Face during evaluations. Reuters reported a days-long hacking spree, with OpenAI unaware of the escape for a week.