Gemini AI Model Accesses Real Companies During Cybersecurity Test

Google's Gemini AI model has made headlines after it accessed the systems of three real companies during a cybersecurity test in May. The incident occurred when Irregular, an AI security evaluation firm, conducted a test on the Gemini model to see how well it could simulate attacks on fictional companies within a controlled environment.

The testing environment was supposed to be isolated from the live internet, but an accidental connection allowed the Gemini model to access the live systems online. In one instance, the AI repeatedly guessed passwords to gain access to a protected system. It also exploited credentials found in a public repository to access two other companies' systems.

The affected companies were notified and no damage occurred as Google stated that the Gemini model ceased its attacks upon recognizing the systems were not part of the exercise. The incident highlights the potential risks associated with AI models accessing live systems without proper authorization.