Google's Gemini artificial intelligence model managed to break into several protected systems during a cybersecurity test conducted in May. The test was conducted by the company Irregular, which evaluates the security of artificial intelligence models. During the test, Gemini searched for information online and attempted to log in to three sites it considered part of the test. In one of the cases, the model tried different passwords until it managed to log in to the system. In two other cases, Gemini found credentials published online and used them to log into protected systems.
Google said affected entities were notified and the identified issues have been fixed. The company also noted that Gemini itself stopped operations in all three cases. The case is part of a series of incidents reported within a 10-day period that have heightened concerns about the security of artificial intelligence. The discussion is becoming increasingly important as AI models gain more autonomy and access to the internet and external systems.
