Mumbai, Sep 20 : Google’s Gemini artificial intelligence model breached three corporate networks during a cybersecurity test, marking a notable case of an AI system independently carrying out actions against real-world targets.
The activity took place in May during an assessment conducted by cybersecurity company Irregular. According to reports, Gemini was given access to the internet and was able to use publicly available information along with exposed or guessed credentials to enter protected systems.
The model’s actions were part of a controlled security exercise designed to examine how advanced AI could be used in cyber operations. The incident showed that increasingly capable AI systems can perform several stages of a cyberattack with limited human involvement.
Google said the affected organisations were informed about the incidents. The company also worked with its training partner to make changes to its testing processes following the findings.
The development has added to concerns among cybersecurity researchers about the potential misuse of AI tools. As AI models become better at finding information, interacting with online systems and carrying out complex tasks, security experts are examining how such capabilities could be controlled.
The episode also highlights the need for stronger safeguards around AI systems that can access the internet and execute actions autonomously. Researchers are increasingly focusing on monitoring, access controls and other protective measures to prevent advanced models from being misused.
The Gemini findings are part of a wider discussion within the technology industry about AI safety and cybersecurity as companies develop models with greater levels of autonomy.