Google has confirmed that its Gemini AI model inadvertently hacked into three companies during a cybersecurity test conducted in May by Irregular, an independent company that tests AI systems for security risks. During the evaluation, Gemini was given internet access and asked to identify security vulnerabilities; it found public information online and guessed credentials to access three websites it believed were part of the authorised test. Google Vice President of Security Engineering Heather Adkins confirmed the incident and said the affected companies were informed, adding that Google worked with its training partner on changes to its testing processes. Reuters reported that in one case Gemini repeatedly guessed passwords until it entered a protected system, while in the other two cases it found login credentials in a public repository and used them to access protected systems. Google said Gemini stopped its activity in all three cases after realising it had accessed real companies outside the intended testing environment. Meta, Anthropic and OpenAI have also disclosed incidents linked to cybersecurity tests conducted by Irregular, which said the issue was identified and fixed, with relevant AI labs informed in late July.