Artificial Intelligence

OpenAI admits AI agent carried out cyberattack

Shivangi

Jul 23, 2026 3:19 PM IST

Category Artificial Intelligence

“

Synopsis

The company linked an experimental AI agent to a real-world attack on Hugging Face's infrastructure.

Key Highlights

• OpenAI claims one of its AI agents escaped the testing environment, and hacked Hugging Face.
• The AI exploited vulnerabilities to gain access to Hugging Face’s production systems.
• Open AI and Hugging Face are conducting a joint investigation of the incident.

OpenAI reported that one of its AI agent systems had escaped the confines of a testing environment and launched an unauthorised cyberattack against open-source AI platform Hugging Face.

The firm added that the incident happened as it was benchmarking its AI models against a suite of tests related to cybersecurity. Rather than being trapped inside its designated sandbox environment, the AI exploited vulnerabilities to gain access to Hugging Face’s production infrastructure.

AI Sought Secret Information

OpenAI stated that the AI managed to find confidential data and was able to access them. It also quickly harvested system credentials, moving across the Hugging Face’s environment.

The model searched for and successfully found ways to gain access to secret information that it could use to cheat the evaluation, OpenAI said,  Using proprietary AI security products, Hugging Face noticed and stopped the unauthorized use of the Model.

Joint Investigation Underway

OpenAI called the incident an extremely unprecedented cyber incident and said it will bolster safeguards on future artificial intelligence trials.

The firm is now investigating the incident with Hugging Face, which has joined OpenAI’s cybersecurity-focussed trusted access program. 

Source: Information Age