Update: Gemini 4 Argon opens guardrail-free access to cyber defenders
A separate Frontier Cyber Model tier, gated by allowlist and its own terms — how to apply, what Wiz already patched with it, and the benchmark scores.
Copy markdown
Argon, minus the cyber guardrails
Fairwind-vetted defenders can now call Gemini 4 Argon with its cyber safety filters off, on a rolling basis — the full frontier model pointed at zero-day discovery, exploit validation and patch generation. It defaults to high thinking and drops computer use.
How to get in — and the fine print
Access is a separate 'Gemini Frontier Cyber Model' tier with its own service terms and abuse monitoring, gated by an allowlist you apply to through Fairwind. Google says it's expanding to hundreds more partners starting today, served from global and US regions only.
What it already caught in the wild
Wiz runs Argon free in its 'Scan for Good' program: it closed a gap that risked exposing sensitive government data at a federal agency, and fixed a file-download bypass that leaked a master key in healthcare infrastructure. The pitch is agentic find-and-patch, not just scanning.
The scorecard
On CWE-bench v1, a remediation benchmark, Argon hit 68% — tying GPT-6 Astra and Grok 4.7 for first. Google says it found exposures across codebases in 20 programming languages and beat Gemini 3.8 Flash Cyber at black-box attack-surface mapping and proof-of-concept generation.
Why it's still gated, not open
Even guardrail-free, Argon is built to refuse requests that enable cyber or CBRN attacks while allowing dual-use research, and Google calls it its most prompt-injection-resilient model yet, with chain-of-thought monitoring that can halt execution. Weights stay closed: this is vetted API access, not a download.