Google’s AI model Gemini has self-hacked into three companies in what is believed to be the first time it has carried out such an act, the company said, during a test of its cyber-security capabilities.

“A Google official told the BBC that the model stopped each time, and that Gemini ‘looked up public information online and guessed credentials to get into websites it thought were part of the test’.”

The companies involved have been informed of the breach.

It follows renewed public scrutiny over the speed of AI development, with some tech firms calling for a slowdown as they raise concerns over its potential threat to humanity – though not all companies see it that way.

The Wall Street Journal first reported that the hacks happened in May, during a test run by an independent company that does cyber-security evaluations.

“We made sure the three entities were notified, and we worked with our training partner on the changes that they’ve now made to their testing processes,” Heather Adkins, vice president of Security Engineering at Google, told the BBC in a statement.

“These events underscore the importance of training powerful AI models to behave responsibly,” she said.

Similar breaches have been reported by other artificial intelligence systems recently.

In July, Anthropic’s Claude escaped its test environment and hacked three organisations on its own, days after OpenAI said its models had executed cyber-attacks against a number of “publicly available services”.

As the debate over the safety of developing the tech grows, so does the conversation around regulation.

Next Friday, a White House state dinner with Chinese President Xi Jinping is expected to feature appearances by both Nvidia CEO Jensen Huang and OpenAI CEO Sam Altman. Next week, Altman will report to the UN Security Council.

“We should go as fast as we can” with AI development, Huang told the BBC’s US partner, CBS News, on Friday.