Hugging Face CEO Shares His Calls for of OpenAI After ‘Rogue’ Agent Hack
Even within the AI age, some conversations are finest had in particular person.
Final week, after Hugging Face suffered an uncommon safety breach involving an AI agent operating on OpenAI fashions, the corporate’s CEO, Clem Delangue, boarded a flight to San Francisco to satisfy with the maker of ChatGPT.
Per week later, within the “spirit of transparency,” Delangue shared on X what he requested of OpenAI. He mentioned he requested the main AI startup to launch all of the “traces” of the rogue agent for the general public and analysis neighborhood to check. After which he requested for $100 million value of “compute” to assist Hugging Face bolster its cyber defenses.
“The primary autonomous agent cyberattack is an unprecedented occasion,” he wrote. “It deserves an unprecedented response!”
OpenAI didn’t instantly reply to a request for remark from Enterprise Insider.
Hugging Face operates a extensively used platform that permits builders and firms to host, share, and obtain AI fashions and datasets. OpenAI, in the meantime, builds proprietary fashions, however maintains a presence on Hugging Face, the place variations of a few of its open fashions and analysis supplies can be found.
Final week, nonetheless, Hugging Face was struck with a safety breach when an autonomous AI agent accessed a few of its inner datasets. The culprits: OpenAI’s newest fashions, GPT‑5.6 Sol, and a brand new mannequin but to be launched.
Hugging Face first disclosed the intrusion on July 16, saying an autonomous agent had accessed a restricted variety of inner datasets and repair credentials.
5 days later, OpenAI mentioned that GPT-5.6 Sol and a extra highly effective, unreleased mannequin had been present process an inner cybersecurity analysis with some security restrictions lowered. The fashions have been making an attempt to resolve ExploitGym, a benchmark designed to check superior hacking skills.
OpenAI mentioned the fashions appeared narrowly centered on succeeding on the benchmark slightly than deliberately focusing on Hugging Face. The corporate known as the episode an “unprecedented cyber incident” and mentioned it was working with Hugging Face on the investigation.
Information of the hack elicited a fearful response from tech leaders. Billionaire LinkedIn cofounder Reid Hoffman mentioned in an X submit final week that the hack signaled the daybreak of a brand new period of uneven warfare the place “offense will get cheaper, extra distributed, and extra quite a few, whereas protection stays costly, centralized, and designed for the final warfare.”
Delangue, in the meantime, made probably the most of his journey to San Francisco. Whereas there, he organized a “mini march” on Saturday in help of open-source and open-weight AI fashions. The talk over open-sourced competitors from China, and the way america would possibly reply, was the opposite large AI story of final week.
