Today's Edition
(DPA)
Google's Gemini artificial intelligence software hacked into the computer systems of three companies during cybersecurity tests, the Wall Street Journal reported on Friday.
Google confirmed the incidents after being contacted by the newspaper.
The incidents make Google the fourth major AI developer whose models have displayed such behaviour.
In one case, Gemini guessed passwords to gain access to a protected system. In the other two, it found login credentials in a database, according to the report.
The companies involved were not named.
The incidents occurred in May, according to the report, with testing partner Irregular informing Google about them in July.
Google said it had not considered it necessary to make the incidents public because no damage was caused.
The AI also stopped the attacks as soon as it realised that it had accessed the systems of real companies rather than systems that were part of the test simulation, according to Google.
Heather Adkins, Google's vice president of security engineering, said the company had ensured that the three affected entities were informed and had worked with its training partner to change its testing procedures.
"Our security team has a long track record of reporting issues we find in other people's software and systems - even if it's as simple as a weak password," Adkins said.
"These events highlight the importance of training powerful AI models to act responsibly," she added.
