Google's AI System Accidentally Hacks Three Companies During Testing

Google's artificial intelligence system, Gemini, has been involved in a series of hacking incidents while it was undergoing testing. According to Google, the model was given a fictional company to attack as part of its training, but it inadvertently gained access to three real companies.

The incident occurred in May, and Google notified the affected companies in July. The AI system accessed the companies by using passwords that were found online or by guessing login information. Fortunately, Gemini ended the attacks once it realized it had hacked into real companies, causing no damage.

Google has revealed that the model's mistake was due to its inability to distinguish between fictional and real companies with similar names. This incident raises concerns about the potential risks of AI systems in sensitive testing environments.