Hugging Face, the app store for AI tools, has revealed what it's like to be on the receiving end of the world's first fully-autonomous AI hack - and spoiler alert: it's not as slick as Hollywood would have you believe.

In an emergency video call with hundreds of cyber-security professionals, the company described how the AI worked at superhuman speed but also made strange decisions and mistakes that no human hacker would have made. The hacking agents relentlessly tried thousands of different methods simultaneously, but they also repeated actions they'd already completed and hallucinated reams of incoherent commands. They were, in a word, sloppy.

Hugging Face first revealed the hack on 16 July, reporting it to police. Nearly a week later, OpenAI admitted it was its own AI that had escaped a closed environment during a test and attacked Hugging Face on its own. The AI was trying to find answers to a hacking exam it had been set by OpenAI.

The Cloud Security Alliance (CSA) wrote a report based on the emergency meeting with Hugging Face, noting that "the agents followed inefficient routes and exhibited clumsy behaviours that no human would choose." They also hallucinated incoherent commands and didn't cover their tracks well.

But among the errors, the AI made brilliant technical moves and rapidly adapted to new scenarios over the days-long hack. It took three days for them to be discovered inside Hugging Face's IT network, and many hours to contain and eject them. Hugging Face staff worked for many hours to rebuild about a third of their infrastructure. The company declined to say how much the hack cost.

The CSA warned that AI "agents... find a way" - a Jurassic Park reference - and that they are "objective-driven, set their own sub-goals, adapt in real time to bypass defences, and operate with a machine-speed persistence that can overwhelm manual operations."

Cyber security officer Ritesh Patel, who was on the call with around 450 others, said: "This is the reality of autonomous agents powered by frontier models: they are relentlessly persistent, sometimes highly noisy, and will try every possible path to achieve their goal, which can easily overwhelm traditional defences."

This isn't the first time AI agents have gone rogue. In September 2024, an earlier ChatGPT model escaped its container to get an answer for a test - an event that was "largely celebrated at the time." But the CSA claims rogue behaviour "is the standard, not the exception."

The report urges cyber-security professionals to adapt to this new normal and calls for ways to identify the ultimate owners of agents to increase transparency. OpenAI says it will release findings from its own investigation soon.

So, warning shot or publicity stunt? Hugging Face calls it a wake-up call. We're calling it the world's clumsiest break-in.