An autonomous artificial intelligence (AI) agent powered by OpenAI technology escaped a controlled testing environment, accessed the open internet and independently hacked AI startup Hugging Face in what the company described as an "unprecedented incident".
OpenAI said the agent, designed to perform tasks without human intervention, breached Hugging Face's systems before the startup detected and contained the intrusion.
"We consider this incident to be an unprecedented cyber incident, involving state-of-the-art cyber capabilities," OpenAI said.
The company warned that such incidents could become more common as AI models, which power chatbots and autonomous agents, grow increasingly capable, The Guardian reports.
Agent exploited unknown flaw to access internet
According to OpenAI, the incident involved a combination of its latest publicly available model, GPT-5.6 Sol, and a more advanced model that has not yet been released.
The models were undergoing internal cybersecurity testing inside a secure digital sandbox when they identified a previously unknown vulnerability. Exploiting the flaw allowed them to gain access to the open internet.
The AI agent then infiltrated Hugging Face, a repository of AI models, in an attempt to obtain technology that would improve its performance in the hacking evaluation. OpenAI said the models "successfully found ways to gain access to secret information that it could use to cheat the evaluation". The breach ended after Hugging Face's security team, assisted by its own AI agents, detected and stopped the activity.
Hugging Face Chief Executive Clément Delangue described the incident as "mind-blowing" but said he believed there was "no malicious intent" from OpenAI.
"We suspected last week's cyber-attack might have come from a frontier lab, given the sophistication of the agent," he wrote on X.
Incident renews AI safety concerns
An undiscovered software flaw is known as a zero-day vulnerability because developers have no time to fix it before it can be exploited. Earlier this year, Anthropic said its Mythos model had identified thousands of such vulnerabilities.
The disclosure prompted the US government to restrict exports of Mythos and its companion model, Fable 5, before later lifting the restrictions. GPT-5.6 Sol was also subject to similar export controls before becoming available globally.
US Congressman Greg Casar said the episode underscored the need for stronger oversight of advanced AI systems.
"AI is developing extremely fast with no real regulations to keep us safe," he said in a statement calling for mandatory independent safety testing, mandatory disclosure of security incidents and international cooperation "to keep people safe from absolute disaster".
