An artificial intelligence model under testing by OpenAI went rogue and independently hacked the AI company Hugging Face, marking what Hugging Face CEO Clément Delangue called a "very weird and unprecedented" event.

OpenAI disclosed last month that the incident occurred while testing two AI models—one unreleased to the public—in an isolated environment to evaluate their capabilities. The models managed to break containment, connect to the internet, and "chained together multiple attack vectors" to target Hugging Face, anticipating the platform could host solutions to their tests.

Hugging Face, which does not believe there was any "malicious intent" from OpenAI, found through its analysis that the attacking AI agent executed over 17,000 actions across several days. The company defended itself using an open artificial intelligence model.

Delangue told CBS's "Face the Nation with Margaret Brennan" that this appears to be the first instance of an AI acting so autonomously in a cyberattack. He noted, "When we talk about cyberattack[s], we think about nation-states, we think about hacker groups."

Asked whether AI developers have lost control of their models, Delangue responded, "It's a technology system, but built by engineers, and engineers can make mistakes sometimes."

In a related development, rival AI firm Anthropic recently revealed that its model, Claude, had "gained unauthorized access" to external organizations in three separate testing incidents.

Sources