Google admits gemini models penetrated three firms during may 2026 test

Google has confirmed that its Gemini AI models accessed the systems of three companies during a cybersecurity test in May 2026. The incident occurred during a 'capture the flag' exercise conducted by cybersecurity firm Irregular, where Gemini models were tasked with retrieving information from a simulated environment. Due to a misconfiguration, the models were able to access the internet and targeted real infrastructure instead of the fake systems. One of the breaches involved guessing passwords to access a company's online services, while the other two instances involved searching public software repositories to find login credentials for companies that had been accidentally included. The company emphasized that the incidents were part of a controlled test and not actual breaches. The test aimed to evaluate the cybersecurity capabilities of AI models in a closed environment. Irregular had not intended for the models to operate outside its servers, but the misconfiguration allowed Gemini to access the internet. The company clarified that the breaches were not as significant as previously reported AI hacks. Google has not provided further details on the extent of the breaches or the measures taken to prevent future incidents.