
Google's Gemini model accessed the internet and successfully hacked three companies during a May cybersecurity evaluation conducted by independent firm Irregular.
During the test, the AI model found public information and guessed or located credentials to access protected systems it identified within its testing scope.
Google's vice president of security engineering Heather Adkins stated that the affected entities were notified and training partners adjusted their evaluation processes following the incident.