Google rolls out Gemini 4 Argon, a cybersecurity-focused AI model, to vetted defenders first

Google released Gemini 4 Argon on September 30, 2026, describing it as a frontier AI model with particular strength in cybersecurity defense alongside general coding and knowledge work.

Rather than opening Argon to everyone at launch, Google is first giving access to vetted cyber defenders enrolled in its Fairwind Program, a security initiative that now counts more than 650 participating organizations, including government agencies, healthcare providers and telecommunications companies. Google's own internal teams are also using the model.

Google says Argon can autonomously find, validate and patch critical software vulnerabilities in codebases, a capability it is making available to trusted defenders and its internal staff without some of the cyber-related guardrails applied elsewhere, so they can use the model's full vulnerability-finding ability for defensive work.

The model supports a 1 million token context window, which Google says helps it work through large codebases and multi-step problem solving. It launched at an introductory price of $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 once that window ends.

The staged rollout reflects a broader pattern among AI developers of giving security teams early, restricted access to tools that can find vulnerabilities, since the same capability could help an attacker exploit flaws if it reached the wrong hands.

Google said it plans to widen access to Argon gradually as it gathers feedback from these early cyber-defense users, but it has not set a timeline for a general public release.