Gemini, Google’s AI model, had reportedly hacked into three different companies during a cybersecurity test back in May, the firm said.

This is the first known case of Google’s AI model carrying out such an attack. The hacking occurrred during cybersecurity testing carried out by Irregular, an independnet cyber evaluation firm.

Irregular was also connected to the Anthropic model hackings seen earlier this year – the firm’s configuration error lead to the model’s breach.

In the case of Google Gemini, the AI model found information published publicly, and guessed the login credentials to access websites the model thought were a part of the test.

Each model stopped at a certain point, a Google official told the BBC, though it remains to be revelaed at what stage each model haulted their actions.

Irregular said that it informed Google and all of the affected firms back in July after its investigation into the incident.

All known issues related to the breach have been resolved, the firm said.

According to reports, the Googel AI model guesss passwords to websites until it gained access to certain systems.

The news comes following reports from OpenAI and Anthropic detailing similar breaches.