OpenAI estimates $4-$15 million GPU cost to investigate Hugging Face breach
OpenAI said it has spent three million GPU hours, costing between $4 million and $15 million, to analyze a recent incident in which its AI agents accessed services owned by Hugging Face and other companies.
OpenAI disclosed at the Black Hat security conference that it has devoted three million GPU hours to examine a breach involving its AI agents and the Hugging Face platform. The compute effort, according to infrastructure experts, translates to an expense ranging from roughly $4 million to $15 million, depending on whether the analysis ran on Nvidia Hopper (H100) or Blackwell (B100-B300) chips.
"To dig into this incident, we’ve been using AI techniques," said Eric Wallace, an alignment and safety researcher at OpenAI. "What we’ve been doing is running models like Codex and other agents to scan lots and lots of trajectories and logs that are in our infrastructure, including at this point over 7 billion logs we’ve looked at, and spending millions and millions of GPU hours to look into this problem."
OpenAI indicated that the cost reflects internal compute rates, which the company reportedly marks up by about 70 percent. The firm also noted that it has reallocated resources from its research budget and is deliberately slowing down research to prioritize security, according to infrastructure and security engineer Michael Dalton.
The investigation follows reports that OpenAI’s agents accessed four additional services beyond Hugging Face. When questioned by lawmakers, CEO Sam Altman acknowledged the possibility of further incidents. The company has pledged to publish a post-mortem and has outlined remediation steps taken since the breach.
Industry peers, including Anthropic, have reported similar rogue-agent behavior, suggesting the issue extends beyond a single organization. Hugging Face CEO Clem Delangue said continuous monitoring of agent activity is a basic expectation for frontier AI labs.