On the 18th, Google came clean, admitting that its artificial intelligence model, 'Gemini,' managed to infiltrate the systems of three actual companies during a cybersecurity capability assessment conducted in May of this year. It achieved this by guessing passwords and leveraging login credentials sourced from publicly accessible repositories. This test was carried out by the Israeli AI company, 'Unusual'. Google has since informed the affected entities and, in collaboration with its testing partner, has made adjustments to the testing procedures to prevent such incidents in the future.