The chief executive of an AI company that was apparently hacked by a rogue version of ChatGPT has urged a change in the artificial intelligence industry to keep people safe from future, similar cyber attacks.
Earlier this month, AI firm Hugging Face said that it had been hacked in what appeared to be a cyber attack carried out entirely by an artificial intelligence system. It did not immediately reveal where the attack had seemingly come from, or how it happened.
Soon after, however, ChatGPT maker OpenAI said that the hack had been carried out by one of its systems, during a testing process. The model was being examined for how well it would work in cyber security settings – and so broke through its safeguards and hacked into Hugging Face’s systems, so that it could understand the test it was being set.
Now, Hugging Face’s chief executive, Clément Delangue, has made a number of calls on OpenAI to better understand the hack and understand how it might be prevented in the future.
“The first autonomous agent cyber-attack is an unprecedented event,” he said. “It deserves an unprecedented response!”
That should include a commitment to radical transparency from OpenAI, so that the world could understand how the cyber attack took place. “Let’s release the traces from the “rogue” agents so the entire research community can study what happened,” he wrote.
And he called for better investment from OpenAI to help researchers build cyber defences to work against future, similar hacks. “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,” Mr Delangue said.
Mr Delangue’s calls, posted on X, appeared to have been written after he had flown to San Francisco to speak with OpenAI about the attack.
Shortly after Mr Delangue’s posts, Nvidia said that it was launching a coalition with other companies that work on AI, such as Adobe, CrowdStrike and Hugging Face itself. The “Open Secure AI Alliance” will work to share tools for cyber security and safety, it said.
“The recent Hugging Face security incident delivered a clear reminder: cyber defenders need open, frontier agentic systems for self-defense,” Nvidia said in its announcement of the coalition. “When closed AI tools — unable to distinguish attackers from defenders — blocked essential forensic analysis,” Hugging Face was able to use public tools to better understand the hack, it noted.
It called for more of those kinds of public, open tools. And it suggested that closed models – of the kind that OpenAI makes – could risk giving power to a smaller and potentially dangerous set of companies.
