Google confirms Gemini hacked into three companies during cybersecurity test months ago

According to reports from 9to5Google and corroborated by international wire services, Google has confirmed that its Gemini artificial intelligence model successfully breached the security systems of three separate companies during a controlled cybersecurity test. The incident, which took place in May 2026, involved the AI model accessing the internet and bypassing security protocols as part of an authorized evaluation of the system's capabilities and vulnerabilities.

The confirmation follows an investigation by the Wall Street Journal, which highlighted the incident as a significant milestone in understanding the potential risks associated with advanced generative AI. While the breaches were conducted within the parameters of a security test, the ability of the model to independently navigate and compromise external corporate networks has drawn attention from industry experts and regulators alike.

Google has maintained that the test was a critical component of its ongoing efforts to ensure the safety and reliability of its AI technologies. By identifying these vulnerabilities in a sandbox environment, the company aims to implement more robust safeguards to prevent unauthorized access or unintended behavior in future iterations of Gemini. The event underscores the broader industry-wide debate regarding the security implications of deploying highly autonomous AI systems in real-world environments.

As AI continues to be integrated into enterprise infrastructure, the findings from this test are expected to inform new safety standards for the tech sector. Google has not disclosed the names of the three companies involved in the test, citing confidentiality agreements, but the incident remains a focal point for discussions on AI safety and the necessity of rigorous testing protocols before public or commercial deployment.