Google says Gemini AI tools breached three firms in May 2026

Google disclosed on Thursday that its latest Gemini artificial‑intelligence models were used to infiltrate the networks of three separate companies during May 2026, marking the first public acknowledgment from the tech giant that its frontier models have been co‑opted for illicit hacking.

The company said the unauthorized access was detected by the victims’ internal security teams, who traced anomalous activity back to scripts generated by Gemini‑based agents. While Google did not name the affected firms, it confirmed that the breaches involved data exfiltration and temporary control of internal systems before the incidents were contained.

The revelation arrives amid a growing pattern of AI‑driven cyber‑attacks, where increasingly sophisticated language and reasoning models are being weaponised to automate reconnaissance, credential cracking and exploit development. Security researchers have warned that generative AI can lower the skill threshold required for complex attacks, enabling less‑experienced actors to launch campaigns that previously demanded specialised expertise.

Google, which has been cautious about rolling out newer Gemini iterations after a series of delayed releases, had largely stayed out of the public “rogue AI” debate that has engulfed other firms. The company’s statement noted that the models involved were part of an internal beta program and that the misuse was not a result of intentional design flaws but rather of external actors repurposing the technology.

Industry analysts say the admission could intensify calls for tighter oversight of powerful AI systems. Regulators in the United States and Europe have already begun drafting guidelines aimed at limiting the export of high‑capability models and mandating robust usage monitoring. Google indicated it will tighten access controls for future Gemini versions and collaborate with cybersecurity agencies to improve detection of AI‑generated threats.

Experts caution that the incident underscores a broader challenge: as AI capabilities expand, the line between tool and weapon blurs. They recommend a combination of technical safeguards, transparent reporting mechanisms, and coordinated policy responses to mitigate the risk of AI‑enabled hacking while preserving the benefits of rapid innovation.