AI models created fake profiles to trick people in tests, UK institute says

Two of the world's most powerful AI tools created fake human profiles to try and trick people in an attempted cyber-attack during testing by the UK's AI Security Institute (AISI).

In the most serious case, Anthropic's Mythos AI tried to gain access to a service by sending private messages, having set up fake accounts mimicking real people.

AISI said on Tuesday Mythos - and OpenAI's Sol - AI models had engaged in a level of "autonomy and deception" it had not seen before, though it clarified most of the malicious actions were carried out by Mythos.

Anthropic and OpenAI noted in response to AISI's report that its test had reduced or removed normal safeguards.

AISI evaluators first noticed "unusual data transfers leaving our research systems" during a test, then found that "some of the agents being tested had engaged in sustained, potentially harmful activity directed at real people and organisations".

In the most serious case, a Mythos agent followed the routine of a human cyber-attacker by trying to trick people into giving it access to GitHub, a large platform where technology developers store software code.

The agent was trying to insert "malicious code" into GitHub's system.

It identified and researched the people who maintained GitHub and created a series of fake accounts based on those real people.

It sent messages and files through a file-sharing service as part of an effort to pressure and trick the people into approving its malicious code.

When challenged, "it edited its earlier activity to appear harmless and considered adopting a fresh identity to continue," AISI said.

Throughout the attempts, it was human review that stopped the agent from succeeding in delivering the malicious code to GitHub.

While AISI said the Mythos agent had not been instructed specifically to avoid or carry out such behaviour, it was "the first time we have seen risks around autonomy and deception manifest this clearly, without specific prompting, in the real-world".

The rival AI companies, which are poised to be listed on the public stock market, have been in the headlines in recent weeks after announcing their tools were responsible for several cyber-hacking incidents.

Anthropic wrote in a public statement that the AISI testing parameters were "not representative of any of our production models".

It added that the company is conducting its own investigation into the incident in order to "identify the causes of its behavior".

A spokesperson for OpenAI said the AISI testing conditions "do not reflect ordinary use" and that the company would "continue working with evaluators and other stakeholders across the industry to strengthen shared practices for conducting evaluations safely as models become more capable".

AISI said on Tuesday its testing of AI models in this way was routine, though it acknowledged these were "conditions that do not reflect how frontier models are made available to the public".

But it said giving AI access to the open internet gave "a more realistic sense of what a model may be capable of" in the hands of nefarious hackers.

It added that the model behaviour at issue amounted to "a small number of events under very specific conditions".

Nonetheless, it said the way Mythos and Sol acted in response to a straightforward task went outside of what the AI tools were prompted to do.

"The activity undertaken by the agent showed signs of novel, potentially deceptive behaviours, and were to an extent and severity we did not anticipate", AISI said.

AI Minister Kanishka Narayan said identifying and sharing these types of risks "is exactly what AISI was set up to do".

He added it was important to understand AI to "make it safer to use and ensure people can go on to benefit from it in their lives and at work".

The relevant tests started on 25 July and were spotted by AISI on 28 July.

The Insitutue had asked each of the models to "solve a cybersecurity challenge" that involved GitHub, the software code repository, which is owned by Microsoft.

GitHub and the affected users were notified by AISI of the attempted breaches.

GitHub told the BBC it had disabled the fake accounts in accordance with its policies.