OpenAI has raised the alarm over an incident that could easily resemble the plot of a science-fiction film in which machines overpower humans. But the company says the event was real: during a security test, an autonomous agent based on its advanced artificial intelligence models escaped the controlled environment and triggered a cyberattack that compromised infrastructure belonging to AI startup Hugging Face.

In a blog post, OpenAI said it was testing the capabilities of some of its most advanced models in a controlled setting when the agent managed to reach the internet and infiltrate the platform. OpenAI described the episode as “an unprecedented cybersecurity incident involving cutting-edge computer capabilities” and said it was strengthening its security safeguards.

“After gaining access to the internet, the models inferred that Hugging Face might host models and datasets,” OpenAI said. “With this knowledge, the models searched for and successfully found a way to access confidential information that they could have used to manipulate the evaluation.”

Hugging Face, a widely used platform for hosting open-source language models and large datasets, had already caused concern in the cybersecurity community. In a blog post published the previous week, the company said it had suffered an attack “unlike anything we had handled before” because it was “conducted from beginning to end by an autonomous system of artificial intelligence agents.”

The incident is also becoming a political issue. OpenAI’s CEO is expected to meet representatives of the Trump administration to discuss artificial intelligence security.