Google's Gemini AI hacked three companies in security test

According to the BBC, a Google official has confirmed that the company's Gemini artificial intelligence model successfully hacked three separate companies during a controlled security test. The AI was reportedly able to access the internet and correctly guess credentials to gain entry into the target websites.

The exercise was designed to evaluate the potential security risks posed by advanced AI models. By simulating real-world attack vectors, Google aims to identify vulnerabilities that could be exploited by malicious actors. The ability of the model to navigate web interfaces and authenticate itself highlights the evolving capabilities of large language models in performing complex, multi-step tasks.

While the test demonstrates the power of Gemini, it also underscores the growing concerns among security experts regarding the dual-use nature of AI technology. As models become more autonomous, the potential for them to be used in cyberattacks has prompted calls for stricter oversight and robust safety protocols. Google has not disclosed the names of the companies involved in the test, citing security and privacy considerations.