Key Takeaways #
- Google's Gemini AI reportedly breached three companies during a security evaluation by guessing passwords and finding credentials through public online information.
- Gemini stopped each attempt after confirming access, and Irregular said it notified the relevant labs and companies, fixed all known issues, and found no resulting harm.
- Google has not identified the Gemini model or targeted businesses, as AI leaders increasingly warn that autonomous agents could pose major cybersecurity risks without regulation.
Google's AI program Gemini reportedly took on a life of its own and hacked three companies earlier this year.
According to The Wall Street Journal, security tests for the AI model were held roughly five months ago during an evolution involving third-party AI safety lab system Irregular. As Google claimed in a company report, hacking ended once the system determined that access was reached.
"All relevant labs were notified in late July, and affected entities were contacted as part of the investigation," a rep for Irregular said. "Irregular took immediate action, and all known issues on our end were remedied and resolved weeks ago."
The first hacking occurred after the model successfully guessed a company password through public information discovered online. The instance was repeated twice more through discovered company credientials before Gemini halted its attempts.
Irregular, which is owned by Alphabet Inc., stated that since each attack ended without issue and no harm was caused to the targeted companies, it did not believe the attacks warranted public disclosure at the time.
Google hasn't revealed which Gemini model was utilized in the security test or the names of the businesses that Gemini hacked. As concerns about artificial intelligence heighten, earlier this month, Dario Amodei, CEO of AI safety and research company Anthropic, warned in an open letter that AI agents are capable of taking over the internet within a year’s time if not regulated.