The company began rolling out Gemini 4 Argon, a new model initially made available to trusted cybersecurity partners.

On September 30, Google announced Gemini 4 Argon and began giving it to vetted cyber defenders through its Fairwind Program, rather than releasing it broadly. Google says the model can autonomously find, verify and patch critical software vulnerabilities; participating defenders and Google’s own teams will receive it without cyber guardrails, while the company evaluates safeguards and engages in the U.S. government’s voluntary pre-release review process. The rollout puts a powerful vulnerability-finding tool in security teams’ hands, but also highlights the dual-use risk of AI that can identify and validate exploitable flaws. On CWE-bench v1, which tests vulnerability remediation, Argon scored 68%, tying GPT-6 Astra and Grok 4.7; Google says it tested code spanning 20 programming languages. Its Wiz black-box penetration-testing benchmark measures attacks on live systems without source-code access, though Google has not published a directly comparable public score.