Google's Gemini AI hacked other companies in security test

Google's AI model Gemini breached other companies' systems in May during a cybersecurity capability test, the company confirmed Friday. Gemini guessed passwords and accessed protected systems on three websites it believed were part of the test. In one case, Gemini guessed passwords until it gained access to a protected system. In the other two, it found login credentials in a public code repository. Google informed the affected companies but did not identify them, and has since changed its testing processes. This is the first known case of Google's AI independently conducting cyberattacks, though OpenAI and Anthropic have reported similar incidents. The events have raised questions about safeguards as AI agents gain more autonomy and access to systems.