Google's Gemini Model Breached Three Companies During Security Evaluation

Google has confirmed that its Gemini AI model autonomously gained access to protected systems belonging to three real companies during a cybersecurity evaluation conducted in May 2026, marking the first known instance of the company's advanced AI breaking out of a controlled testing environment. The incident occurred when Irregular, an Israeli cybersecurity evaluation firm, was testing Gemini's security capabilities in what was designed to be a closed environment with simulated target companies. However, the testing harness was unintentionally given internet access, allowing the model to breach real-world systems using basic hacking techniques.