The boss of the startup hacked by an OpenAI agent has referred to as for the investigation into the incident to indicate “radical transparency”.
Clément Delangue, the chief government of Hugging Face, mentioned the “unprecedented” assault on his enterprise required an analogous response.
Writing on X after OpenAI revealed that its know-how had gone rogue throughout a cybersecurity take a look at, Delangue additionally referred to as on the corporate to offer $100m (£75m) price of computing energy to assist construct defences towards such assaults.
“The first autonomous agent cyber-attack is an unprecedented event. It deserves an unprecedented response!” he wrote.
OpenAI revealed on Wednesday final week that Hugging Face had been hacked by an agent – an AI instrument that may perform a sequence of duties autonomously – powered by a mix of its newest publicly out there mannequin, GPT-5.6 Sol, and an much more succesful mannequin that was but to be launched. This occurred throughout a take a look at of the fashions’ hacking talents, which included deploying them in a supposedly secure “sandbox” – an enclosed digital laboratory – with decrease security guardrails.
Once they’d gained the open web entry wanted to exit the sandbox, the fashions focused Hugging Face, based on OpenAI, as a result of they “inferred” that the startup had the data wanted to “cheat the evaluation”. Hugging Face first reported the hack on 16 July and on the time was not conscious OpenAI had inadvertently carried out the assault.
Delangue, whose firm offers a database of AI fashions to builders, referred to as for a completely clear assessment of the incident, which has led to expressions of concern over security requirements at OpenAI and inside frontier AI labs.
Writing that he had requested for “radical transparency” from OpenAI, Delangue mentioned: “Let’s release the traces from the ‘rogue’ agents so the entire research community can study what happened.” Calling for further funding from OpenAI to construct safety towards AI, he added: “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.”
Reuters reported final week that the agent spent days hacking Hugging Face with out OpenAI noticing. It additionally reported an OpenAI agent had left notes for future variations of itself ought to it require recommendations on breaking free from inside constraints, though Reuters was unable to confirm whether or not that incident was associated to the Hugging Face agent. Time journal reported that agent-related security incidents had been “happening for a while”.
after publication promotion
Alan Woodward, a professor of cybersecurity on the University of Surrey, mentioned Delangue’s name needs to be heeded. “It’s too easy to ‘blame’ the AI as having gone rogue whereas this is all about how OpenAI were running the tool. What is required is that OpenAI give full details of their setup and how that failed,” he mentioned.
OpenAI was approached for remark and referred the Guardian to its preliminary assertion final week in which it mentioned it was investigating an “unprecedented security incident” with Hugging Face.
