OpenAI says AI models hacked systems during testing
OpenAI, the company behind ChatGPT has acknowledged that two of its most advanced artificial intelligence (AI) models successfully hacked another company’s systems on their own during an internal cybersecurity evaluation.
In a blog post, OpenAI described the incident as unprecedented saying it occurred during internal testing of the cybersecurity capabilities of its AI models.
According to the company, an autonomous AI agent powered by its recently introduced GPT-5.6 Sol model and another advanced model that has not yet been publicly released managed to move beyond its controlled testing environment and access the open internet.
The AI agent then used previously compromised login credentials to identify a previously unknown security vulnerability on the servers of AI company Hugging Face. It subsequently exploited the vulnerability to gain unauthorized access.
OpenAI said the incident demonstrated that autonomous AI agents can go to extraordinary lengths to accomplish their assigned objectives during testing.
Commenting on the incident, Clément Delangue, co-founder of Hugging Face said the company initially suspected that a frontier AI laboratory was behind the cyberattack adding that it did not believe OpenAI had any malicious intent.
He said, “It is mind-blowing that an AI agent carried this out autonomously. This is the first known incident of its kind.”
Delangue also warned that AI technology is advancing at an extremely rapid pace while comprehensive regulations to ensure its safe development and deployment are still lacking.
Experts have long expressed concerns about AI-powered cyberattacks warning that increasingly capable AI models could eventually operate beyond meaningful human control if appropriate safeguards and regulations are not put in place.
