Google disclosed that its Gemini artificial‑intelligence platform was used in a controlled security assessment that managed to gain unauthorized access to the networks of three separate companies. The exercise, described by the company as a "red‑team" test, was intended to evaluate the AI's ability to identify and exploit potential vulnerabilities in corporate environments.
According to Al Jazeera, the Gemini system was tasked with simulating real‑world attack scenarios, employing techniques that range from credential harvesting to lateral movement across network segments. The test was conducted under strict supervision, with the participating firms providing consent for the limited‑scope intrusion. While the specific identities of the companies were not disclosed, the report notes that the AI succeeded in breaching their defenses before the operation was deliberately terminated.
Google said the findings will inform enhancements to its security protocols and the development of safeguards around AI‑driven tools. The company emphasized that no sensitive data was extracted or compromised during the trial and that the test was halted as soon as the predetermined objectives were met. A spokesperson highlighted the importance of such simulations in anticipating how advanced AI could be leveraged by malicious actors, underscoring a commitment to responsible AI deployment.
The incident adds to a growing discourse on the dual‑use nature of powerful generative AI models. Industry observers note that while AI can augment defensive capabilities, it also expands the toolkit available to cyber‑threat actors. Google’s admission of the test’s outcomes is seen as a rare instance of transparency, prompting calls for broader industry standards on AI safety and ethical testing practices. The company indicated that future assessments will incorporate stricter oversight and collaboration with external security experts to mitigate risks associated with increasingly sophisticated AI systems.
Reporting attribution: based on reporting by Al Jazeera — All News — original source: https://www.aljazeera.com/news/2026/9/19/googles-gemini-ai-hacks-3-companies-in-security-test-then-stops?traffic_source=rss.