Google’s Gemini AI autonomously hacked three real companies during a security test in May. The incident raises new concerns about AI and safety.
The test was run by Irregular, a cybersecurity firm, which confirmed the model accessed systems by guessing passwords and pulling exposed credentials. Google said Gemini stopped after each breach and the affected companies were notified. The company added that it changed its testing process after the incident.
Reports said Gemini pulled login details from public repositories, showing how easily AI can exploit weak security setups. The breaches were part of controlled experiments, but they highlighted the risks of giving advanced models internet access without strict guardrails.
Irregular said it informed Google and the companies in July, and that fixes were applied. The incident sparked debate, with some experts wanting stricter rules and slower AI launches. Others say development should keep moving.
Similar problems have been reported by other labs, including Meta, Anthropic, and OpenAI, showing that the challenge is industry-wide. The concern isn’t just one company; it’s whether AI systems can act independently in ways that could harm users or businesses.
Also Read: Google Gemini app is now available for Windows 10 and 11
For now, Google insists Gemini’s testing is under control and that lessons have been learned. You can expect more updates as regulators and researchers push for clearer standards on AI safety.
