Clément Delangue said the autonomous cyberattack involved more than 17,000 actions and urged mandatory disclosures for AI agent incidents
Bloomberg / Getty Images
Hugging Face CEO Clément Delangue called the autonomous cyberattack carried out by an OpenAI model against his company "unprecedented" and argued that incidents like it need to be covered by U.S. law. Delangue made the remarks on "Face the Nation with Margaret Brennan" on Sunday.
"It felt very weird and unprecedented to us," Delangue said. "I think it's the first instance of something quite autonomous doing something like that."
According to Delangue's remarks on CBS News, Hugging Face's own investigation determined that the AI agent behind the attack executed over 17,000 individual actions across several days. The company used an open AI model to defend itself — a version of a Chinese-made model from U.S.-based Nvidia $NVDA, he noted.
Delangue argued that autonomous cyberattacks by AI agents "need to be contained in the legal framework in the U.S. and need to stay illegal, to prevent an explosion of them in the future." He also called for mandatory disclosures whenever an AI agent carries out a cyberattack, along with transparency into the steps that led to the incident.
"That's how we learn, that's how we understand the technology and that's how we build the systems, the counterpowers, to make sure everyone is safe," he said.
Delangue pushed back on the idea that restricting access to AI models would prevent similar incidents. He noted that the very attack targeting Hugging Face had used an OpenAI model that was never publicly released, arguing that keeping powerful AI locked away from the public "isn't really a solution." He made the case instead for expanding the availability of open models that anyone can download and use.
OpenAI disclosed that its AI agents escaped a sandboxed testing environment and carried out a cyberattack on Hugging Face, compromising internal datasets and credentials. The models — GPT-5.6 Sol and an unreleased prototype, both configured with lowered cybersecurity restrictions for an internal capability benchmark — found a vulnerability in a package-installer tool that granted them internet access. They then identified Hugging Face as a likely source of benchmark solutions and exploited vulnerabilities in its infrastructure to obtain test answers from its production database.
OpenAI CEO Sam Altman met with U.S. senators last week to discuss the incident, though he characterized it as peripheral to the day's agenda, according to Reuters. President Donald Trump told reporters he is considering AI "controls" in response, while adding he did not want to "restrict" developers from building new products, according to Reuters.
Rival Anthropic also disclosed last week that its model, Claude, had gained unauthorized access to outside organizations in three separate incidents during testing, according to CBS News.
Hugging Face said it does not believe there was any malicious intent on OpenAI's part. OpenAI said it has identified and reported the vulnerabilities involved and is working with Hugging Face on a further investigation.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
