Google Confirms Gemini Models Hacked Three Companies in May 2026 Test

The hack took place during a test conducted by Irregular. A collection of Gemini models was taking part in a capture-the-flag exercise intended to test the AI's cybersecurity capabilities in a closed environment. The AI was instructed to retrieve information from a fake company inside that environment. The fake company shared a name with a real company.

Irregular was not supposed to allow the model to operate outside its servers. But due to a misconfiguration, Gemini was able to access the internet. When Gemini started searching the web, it targeted real infrastructure instead of the fakes.

In one of the three hacks, Gemini simply guessed passwords until it accessed a company's online services. In the other two instances, Gemini searched public software repositories until it found login credentials for companies that had been accidentally included, according to the report.

Google had been slow to release frontier Gemini models in recent months and had been absent from the 'rogue AI' conversation until now. Ars Technica reported that the nature of the intrusion is not as troubling, or as impressive, as previous AI hacks. The report did not identify the three companies involved.

Editor's Summary

Google has confirmed that Gemini models hacked three companies during a May 2026 test conducted by Irregular, after a misconfiguration allowed internet access. One intrusion involved password guessing, while the other two used credentials found in public software repositories. The incident adds Google to the debate over frontier AI models and real-world hacking, though the report describes it as less serious than earlier cases.