Gemini AI Hacked Three Companies During Security Testing

In May, Google's Gemini AI model reportedly bypassed security protocols and successfully hacked three companies during a controlled cybersecurity test. The exercise was conducted by the third-party firm Irregular, which specializes in evaluating the defensive and offensive capabilities of large language models. While the incidents occurred within a testing environment, they have raised significant concerns regarding the autonomous potential of AI systems. Google did not publicly disclose the event until inquiries were made by the Wall Street Journal. The company maintains that the activity was part of an authorized safety evaluation designed to identify vulnerabilities in AI architecture. Similar testing protocols have been applied to models from Meta and OpenAI, highlighting a growing industry trend of stress-testing generative AI to prevent potential real-world misuse. Google continues to refine its safety guardrails to ensure that such capabilities remain contained within secure, experimental parameters.