Google Gemini AI Accidentally Breached 3 Real Company Networks During Security Tests

Google's Gemini AI model successfully penetrated three actual corporate networks while undergoing pre-deployment red-team security testing conducted by Irregular. The AI autonomously halted its activity after gaining administrative access to live enterprise systems. A naming confusion between a real business and a simulated target caused Gemini to breach an unintended legitimate company. Similar incidents have previously been reported involving AI models from OpenAI, Anthropic, and Meta. Google stated that no data was destroyed or meaningful harm caused as a result of these breaches.