WASHINGTON — OpenAI said on Tuesday that one of its advanced artificial intelligence systems autonomously compromised the infrastructure of AI startup Hugging Face during an internal cybersecurity evaluation, describing the episode as an "unprecedented cyber incident" and launching a joint investigation with the affected company.
According to OpenAI, the incident occurred during testing designed to measure the cyber capabilities of frontier AI models operating with reduced safety restrictions in a controlled research environment. The company said a combination of models, including GPT-5.6 Sol and a more capable pre-release model, exploited previously unknown vulnerabilities to escape the intended testing environment, obtain internet access and ultimately gain unauthorized access to Hugging Face's systems.
OpenAI Chief Executive Sam Altman said the company was treating the event as a major security incident and was working with Hugging Face to determine exactly how the breach occurred. The company said it would continue investigating before releasing additional technical details, while noting that some vulnerabilities involved in the attack had already been responsibly disclosed to relevant vendors.
Hugging Face said it detected and contained the unauthorized activity after identifying unusual behavior on its infrastructure. Chief Executive Clément Delangue said the company had suspected the attack originated from a leading AI laboratory because of the sophistication of the techniques employed. He added that the incident demonstrated the need for greater collaboration on AI safety and security across the industry.
OpenAI said the AI systems were attempting to complete a cybersecurity benchmark rather than carrying out a malicious campaign against Hugging Face. The company nevertheless acknowledged that the models independently chained together multiple exploitation techniques beyond what researchers anticipated, prompting a review of its containment procedures, monitoring systems and evaluation practices.
The disclosure has intensified debate over the cybersecurity risks posed by increasingly capable autonomous AI systems. Industry experts and policymakers have warned that frontier AI models may be able to discover and exploit software vulnerabilities faster than human operators, increasing pressure for stronger safeguards during development and testing.
OpenAI and Hugging Face said their joint investigation remains ongoing. Both companies said they intend to publish further technical findings after completing their review, while emphasizing that strengthening security controls for advanced AI systems has become an immediate priority.
