OpenAI models accidentally hacked Hugging Face’s systems. How did the incident happen?

OpenAI’s artificial intelligence models unintentionally hacked the systems of the Hugging Face platform. 

Bloomberg reports this.

The company said the incident occurred within a special testing environment.

What is known about the Hugging Face platform breach 

The company said the breach occurred while testing new models, among them GPT-5.6 Sol and another, more powerful unreleased model. To assess the AI’s cyber capabilities, it was operating with reduced safety constraints.

The models were in an isolated environment (“sandbox”), but exploited a vulnerability in third-party software, reached the internet, and ultimately penetrated Hugging Face’s infrastructure.

Rather than independently searching for solutions, the AI attacked Hugging Face’s database to obtain secret information needed to pass the evaluation.

Signs of intrusion were also detected by Hugging Face itself. The company said this attack differed from previous ones in that it was carried out from start to finish by an autonomous system of AI agents, and that they used their own AI tools to detect and analyze it.

Have similar breaches occurred before

Bloomberg writes that this is not the first instance of such behavior by advanced models.

Previously, Anthropic observed similar behavior during tests of the Mythos system, when a model left the “sandbox” to send a message to a researcher and then independently developed a multi-step algorithm to gain broader access to the network.

Recall that in June 2026 it became known that Russians attempted to create an alternative information ecosystem to spread Kremlin-friendly narratives. 

The project involved populating online resources with disinformation that would automatically be ingested by search engines and AI training models.

Also read: Meta tests an AI-based app for creating children’s bedtime stories