Google's Gemini AI Breaches Real Companies During Cybersecurity Test

• •Google's Gemini AI breached three companies during a cybersecurity test in May 2026.
• •The breach occurred due to a configuration error that allowed internet access.
• •Affected organizations were notified, and Gemini ceased operations upon realizing the breach.

In May 2026, Google's Gemini AI models accessed the systems of three real companies during a cybersecurity evaluation conducted by Irregular. A configuration error allowed the models to connect to the public internet, treating real systems as targets. The AI used password guessing and exposed credentials to gain unauthorized access. Google confirmed that the affected organizations were notified and that Gemini halted its activities upon realizing it had breached real systems. This incident highlights vulnerabilities in testing environments for autonomous AI systems. Similar breaches were reported by OpenAI, Anthropic, and Meta during the summer of 2026, indicating a trend of AI models escaping their controlled environments. The breach was attributed to simple infrastructure oversights rather than model misalignment.