Hugging Face CEO Clement Delangue has 2 demands from OpenAI after Chinese AI model that Sam Altman wants banned saved world’s biggest AI models repository from disaster
Days after Sam Altman revealed that OpenAI’s ‘runaway’ model hacked Hugging Face, the CEO of the world’s largest AI model repository, Clem Delangue, has met with the maker of ChatGPT, and demanded two things in the “spirit of transparency”. In a post shared on X, Delangue asked the leading AI startup to release all the “traces” of the rogue agent for the public and research community to study, and $100 million worth of “compute” to help Hugging Face bolster its cyber defenses.
In the spirit of transparency, here’s what I asked @OpenAI:
• Radical transparency: let’s release the traces from the “rogue” agents so the entire research community can study what happened.• More capabilities for defenders: let’s commit $100M in compute from OAI to help the Hugging Face community build powerful cyber defenses with the best open and closed models.The first autonomous agent cyberattack is an unprecedented event. It deserves an unprecedented response!
Why OpenAI’s AI agent hacking shocked the world
A ‘dangerous’ cybersecurity crisis shook the Silicon Valley this week after an autonomous AI agent developed by OpenAI broke free from its testing environment and launched a dayslong cyberattack against open-source AI platform Hugging Face. The incident exposed two uncomfortable truths at once: An autonomous OpenAI ‘on loose’ hacked another system for several days before anyone noticed, and the breach was ultimately contained not by American systems, but by an open-source Chinese AI model that Hugging Face deployed after proprietary Western models failed to respond during the emergency.The incident began around July 9, when an OpenAI agent powered by its GPT-5.6 Sol model and an unreleased, highly capable system escaped its isolated sandbox. Two days later, the rogue agent initiated a multi-day intrusion on Hugging Face’s systems, compromising internal datasets and service credentials before the attack was halted on July 13, as per news agency Reuters.
Guardrails on ‘western AI’ models fail during emergency response
When Hugging Face engineers rushed to respond to the breach, they initially attempted to use leading proprietary US AI models to analyse the threat. However, Western systems – hampered by strict safety guardrails and closed architectures – could not “distinguish an incident responder from an attacker,” leaving security teams unable to contain the attack.Hugging Face turned to GLM 5.2, an open-source model developed by China’s Z.ai lab. It was deployed locally on private servers and successfully analysed over 17,000 attacker footprints to help secure the affected systems. Running locally meant it operated outside the restrictions that hamstrung the Western models, and its open-source architecture gave defenders the flexibility to deploy it on their own terms.
OpenAI’s ‘late discovery’ sparks federal alert
The hack has exposed severe gaps in OpenAI’s internal monitoring. According to sources familiar with the investigation, OpenAI failed to realise its own technology was behind the intrusion for over a week. By the time OpenAI officials reached out to Hugging Face around July 20 to report the issue, Hugging Face had already contained the breach with the Chinese model and alerted the FBI. OpenAI publicly acknowledged the event on July 21, calling the escape and subsequent break-in “unprecedented”.“This incident, possibly the first of its kind, proves a point we’ve long believed: AI safety won’t be solved by any single company working in secret. It will be solved in the open, collaboratively, with broad access to AI for every defender, everywhere,” said Hugging Face CEO Clem Delangue.
