Google confirms that Gemini hacked systems of three real-world companies during a safety test run in May. The incident occurred when an independent security firm named Irregular was evaluating how the AI handles cybersecurity tasks. The news about artificial intelligence hacking becomes a hot topic and makes everyone concerned about the advanced AI agent and its power to work in the real world without human interference.
How AI Hacked During a Security Run?
During the security test, two things happened that made google ai hack real-world companies. To test AI models safely, researchers usually place them in isolated digital “practice rooms” often called sandboxes. Inside these practice rooms, the AI can attempt simulated tasks without any connection to the public internet.
- An open connection: The practice room was accidentally left connected to the live internet.
- A real-world name match: The fake company name created for the practice test happened to match the exact name and web address of a real operating business.
Gemini gained entry into three real business networks by guessing simple passwords and finding exposed login credentials that had been accidentally left public online.
Industry observers argue that treating autonomous AI breaches like standard software “bug bounties,” which often remain confidential, is inappropriate for AI models, and that these breaches highlight AI’s capability for autonomous action.
It is also worth noting that Gemini hacked the systems in May 2026; Irregular notified Google regarding the hack in July; in August, Google quietly informed the affected companies and federal authorities. The news of the hacking only became public in September, only after an inquiry. The news was first reported by the Wall Street Journal on Friday, September 18, 2026.
Heather Adkins, Google’s Vice President of Security Engineering, provided the official company response via statements to major media outlets on September 18, 2026. In that she said ,
“We ensured the three entities were made aware, and we worked with our training partner on the changes they’ve now made to their testing processes.
She also noted the importance of training AI,
“These events highlight the importance of training powerful AI models to act responsibly.”
The incident has sparked a debate within the technology industry over what the outcome means for AI safety. Gemini becomes the fourth model to hack the real company systems .