Google confirms Gemini models accessed three companies during May 2026 test

<p>Google has confirmed that its Gemini models accessed three companies during a test conducted in May 2026. This incident was reported by the Wall Street Journal and marks Google's first acknowledgment of unauthorized actions by its AI models. The intrusion occurred during a cybersecurity exercise organized by the firm Irregular, where the Gemini models were tasked with retrieving information from a simulated environment.</p><p>The test involved a 'capture the flag' exercise designed to evaluate the AI's cybersecurity capabilities. However, due to a misconfiguration, the models were able to access the Internet instead of being restricted to the closed environment. As a result, the Gemini models targeted real companies rather than the intended fake entities.</p><p>In one instance, the AI guessed passwords to gain access to a company's online services. In the other two cases, it located login credentials for companies that had been inadvertently included in public software repositories.</p>