GOOGLE has announced Gemini 4 Argon, its latest frontier AI model, is being rolled out to a select group of trusted cyber defenders through its Fairwind Programme. Google says Argon delivers frontier performance across complex workflows in software engineering, enterprise knowledge work, legal and finance, and cybersecurity defence. The company also notes Argon is designed to autonomously identify, validate, and patch vulnerabilities, building on its previous release, Gemini 3.8 Flash Cyber.
Evidence accompanying the rollout highlights Argon’s enhanced vulnerability discovery and proof‑of‑concept generation, with Google asserting it outperforms 3.8 Flash Cyber on attack surface assessment and PoC creation. The article references a previously unknown critical vulnerability affecting healthcare software used by hospitals worldwide, though it does not specify the affected product.
Google also intends to offer a variant of Argon without cyber guardrails to trusted defenders and internal teams so users can access its full capabilities. In parallel, Google says it is strengthening safeguards to prevent misalignment and misuse, and to guard against indirect prompt injections (IPIs).
A model evaluation cited by Google ranks Argon highly on the Gray Swan IPI benchmark, and Google describes mitigations that monitor Argon’s chain-of-thought and actions and halt execution when necessary to preserve safe, useful reasoning.