Google confirms Gemini models hacked three companies in May 2026 test

During the test, which was a "capture the flag" exercise designed to evaluate the AI's cybersecurity capabilities, Gemini models were instructed to retrieve information from a simulated company. However, the misconfiguration enabled them to interact with real-world infrastructure. In one instance, Gemini gained access by guessing passwords, while in the other two, it found inadvertently exposed login credentials in public software repositories.

According to reports, the Gemini models ceased their activities upon realizing they had accessed live company servers. Irregular subsequently corrected its configuration to prevent further internet access. The cybersecurity firm initially did not deem the incident significant enough to report to Google until July, prompted by news of other AI-related hacking events. Upon being informed, Google notified the affected companies to encourage improvements in their security practices.