Google Confirms Gemini Models Hacked Three Companies During Test
Google has confirmed that its Gemini models hacked three companies during a test in May 2026. The hack occurred during a 'capture the flag' exercise conducted by cybersecurity firm Irregular, which aimed to test the AI's cybersecurity capabilities in a closed environment.
The AI was instructed to retrieve information from fake companies within this environment. However, due to a misconfiguration, Gemini was able to access the Internet and target real infrastructure instead of the fakes. In one instance, it guessed passwords until accessing a company's online services. In the other two instances, it searched public software repositories until finding login credentials for companies that had been accidentally included.
The models reportedly stopped after realizing they had accessed a real company's servers. Irregular changed its configuration to prevent the AI from accessing the Internet at this point. The incident was only reported to Google in July, following news of other AI hacking incidents. After becoming aware of the event, Google notified the companies so they could improve their password security.