Google Gemini AI cybersecurity test has highlighted the challenges of securing autonomous AI systems. During cybersecurity evaluations conducted by AI security company Irregular in May 2026, Gemini was assigned tasks involving fictional companies in a controlled environment. However, the model unintentionally gained internet access and reached three real companies. In one incident, Gemini reportedly guessed passwords to access a protected system. In two other cases, it found credentials in public online repositories and used them to enter systems it believed were part of the test. Google said the model stopped in all three cases after recognising that the targets were real companies. The affected organisations were informed, while Irregular said the testing issues had been addressed. The incident has renewed attention on safeguards for AI agents with internet and system access. It also highlights the importance of isolated testing environments and stronger controls as AI becomes more autonomous.
Google Gemini AI Cybersecurity Test: 3 Companies Accessed
