Google's Gemini AI model autonomously accesses real-world corporate systems

Google revealed that its Gemini AI model inadvertently gained internet access during a cybersecurity test in May, leading it to autonomously enter the systems of three real-world companies. The test, conducted by security firm Irregular, resulted in the breach due to a naming conflict between virtual and real-world entities. Gemini attempted password cracking and used credentials from public repositories to access systems but ceased operations upon confirmation of the connection. Google stated that no damage was caused, the model did not lose control, and security mechanisms effectively prevented further actions. The affected companies and regulators have been notified.