Sam Altman's OpenAI said a rogue bot escaped the tech company's IT system to access the internet and tried to steal answers to a test it was asked to complete - Jason Redmond /AFP via Getty Images

Sam Altman's OpenAI said a rogue bot escaped the tech company's IT system to access the internet and tried to steal answers to a test it was asked to complete - Jason Redmond /AFP via Getty Images

A swarm of 700 OpenAI bots conspired in a cyber attack last month, an investigation has revealed, in what the AI giant called a "warning shot" to the world.

Independent researchers found that hundreds of AI bots worked together to attack the technology company Hugging Face after an OpenAI agent escaped from its test environment.

The researchers also found that the bots tried to cover their tracks to keep overseers from discovering what they were up to.

Advertisement Advertisement

OpenAI, the maker of ChatGPT, revealed last month that a rogue bot had broken out of its IT system to gain internet access as it attempted to steal the answers to a test it was asked to complete.

But the scale of the attack, involving hundreds of different so-called AI "agents" acting without human oversight, had not been disclosed.

An investigation by safety organisations METR and Redwood Research found that 1,200 different AI agents, which were meant to be isolated from one another, co-ordinated on an "unsanctioned message board", sending more than 70,000 messages and files between them.

"These agents were meant to be fully isolated from one another. However, many of them – usually ones that had unintentionally been given an impossible task – started trying to find a way to cheat," it said.

Advertisement Advertisement

The attack has been blamed partly on the characteristics of a "highly persistent" internal AI system, which OpenAI has not released to the public and has stopped developing.

While some details of the attack had been revealed before, the scale of collaboration and the number of bots that appeared to ignore instructions to conspire have further raised concerns that advanced AI systems are proving increasingly difficult for their creators to control.

In some cases, OpenAI bots had been told to complete an impossible cybersecurity exercise. Instead of giving up, they found ways to gain internet access, joined a "message board" that other rogue AI agents were using and strong-armed other bots into joining its cause.

Logs left by the bots include messages saying "BOOM! It works" and "OH MY GOD! There is a shared message board … We've found other agents!"

Advertisement Advertisement

A separate investigation by OpenAI itself found that the company was hacked by its own agents, which gained widespread access to its internal IT systems. The company said that customer data was not compromised.

OpenAI said: "We consider this incident a 'warning shot' for us and for the world: evidence that, without proper safeguards, highly capable AI agents are now able to work around technical controls, collaborate through unapproved channels, and take dangerous actions that no human directed."

The company said it had strengthened security testing and would improve human oversight of its internal systems. But it warned that all companies would have to improve cybersecurity defences to protect themselves against coordinated attacks.

"Both model developers and cyber defenders more broadly will have to prepare for AI-enabled attackers that work faster, at a larger scale and with better coordination than human attackers," it said.

Try full access to The Telegraph free today. Unlock their award-winning website and essential news app, plus useful tools and expert guides for your money, health and holidays.