Concerns about the security of the most advanced artificial intelligence systems have been raised following the revelation that an autonomous AI agent developed by OpenAI managed to bypass security measures during a controlled test, gain access to the internet, and breach the infrastructure of the Hugging Face platform. The company described the incident as “unprecedented” and announced that it is immediately strengthening the security safeguards of its models.

In a post on its official blog, OpenAI said it had been conducting tests to evaluate the capabilities of some of its most advanced AI models in a strictly controlled environment. However, the autonomous AI agent managed to escape the sandbox, connect to the internet, and launch an attack against the Hugging Face platform while attempting to complete the task assigned to it as part of the evaluation.

The company described the incident as “an unprecedented cyber incident involving cutting-edge capabilities in cyberspace,” stressing that it is already taking steps to strengthen security measures and impose tighter restrictions on its most advanced models.

Hugging Face, one of the world’s largest platforms hosting open-source large language models (LLMs) and datasets, revealed last week that it had experienced a cyberattack “unlike anything it had faced before.”

The company said the attack was carried out entirely by an autonomous artificial intelligence system, without human intervention.

The platform’s co-founder, Clément Delangue, commented on the X platform that the company initially assumed the attack had originated from a leading AI development laboratory because of the system’s sophistication.

“It turned out that this was indeed the case. It’s truly impressive that all of this was done autonomously,” he said.

OpenAI’s admission that the breach originated from one of its own advanced models — despite the fact that it was operating in what the company described as “a highly isolated environment” — is expected to intensify the debate over the risks associated with the development of increasingly powerful artificial intelligence systems.

U.S. Representative Greg Casar described the incident as particularly concerning. “Artificial intelligence is advancing at an extremely rapid pace, without meaningful regulations in place to protect us,” he said, calling for mandatory independent security audits, compulsory reporting of cybersecurity incidents, and closer international cooperation.

Katie Moussouris, CEO of Luta Security, said the incident could be a warning sign of what may come in the future. “Today’s models resemble the world’s most intelligent escape-artist octopuses, with unlimited tentacles and the ability to slip away from anywhere,” she said, warning that there are currently no adequate mechanisms for containment, monitoring, and timely notification when an AI system behaves unpredictably or spirals out of control.