Google unveils Gemini 4 Argon, but only vetted cyber defenders can use it

Google has unveiled Gemini 4 Argon, a frontier model it says is built for long, multi-step work in software engineering, legal and finance tasks, and cyber defence. Ordinary users cannot try it yet. It is rolling out first to trusted cyber defenders through Google DeepMind's Fairwind Program.[1][2]

Google says developers, enterprises and consumers will follow, starting with paid API customers and Google AI Ultra subscribers, but gave no date. Every performance figure is Google's own, apart from a finding it attributes to its partner Wiz, and none has been independently tested.[1]

The Fairwind Program is the only route to Argon at the moment. Google prioritises governments and national cyber authorities, operators of critical infrastructure such as healthcare, telecommunications, energy and financial networks, and core technology platforms. It works with more than 650 partners worldwide.[3]

Access carries conditions. Partners must use phishing-resistant multi-factor authentication, give Argon only to cybersecurity, incident response or penetration testing teams, and track who uses it. Google runs background checks and bars sharing or reselling access.[3]

Google's headline claims are about endurance. It has raised Argon's output limit to 1 million tokens from 64,000, so the model can reason through a very long task in a single attempt. It scores 77.9% on DeepSWE v1.1, a long software engineering test Google calls a new state of the art, and ranks first on Zapier's AutomationBench at 51.3%.[1]

Google also reports 91.7% on LVBench for long-video understanding. Internally, it says Argon agents found memory savings of more than 300 TiB across its data centres and rewrote a video decoder in memory-safe Rust that runs 2.7 times faster than an earlier port.[1]

Cyber defence is where Google is both most ambitious and most careful. It says Argon can find, validate and patch serious software vulnerabilities on its own. On CWE-bench v1, which tests whether a model can fix security flaws, it scores 68% and ties for first place.[1]

Wiz, a cloud security firm, is using Argon in Scan for Good, a free programme that looks for risky exposures in critical public infrastructure. Google says Argon found a critical flaw exposing sensitive personal information across healthcare software used by hospitals worldwide, which earlier frontier models had missed. The announcement names no product or hospital.[1]

Google says it is still strengthening safeguards in four areas before broad release: misuse, prompt injection, misalignment and hardened test sandboxes. Trusted defenders and Google's own teams will get Argon without its cyber guardrails.[1]

It calls Argon its most resilient model yet against indirect prompt injection, where hidden instructions hijack a model. It also monitors the model's chain of thought and actions and can stop execution. Google says it is taking part in the US government's voluntary pre-release access process.[1]

Malaysian teams cannot use Argon yet. Google's Gemini API lists Malaysia among its available regions, but Google has not said whether Argon will be offered here, or whether any Malaysian body has joined Fairwind. At Bank Negara Malaysia's 1 October selling rate of RM4.093 per US dollar, the introductory price is about RM8.19 per million input tokens and RM40.93 for output.[7][1][8]

The nearer link is regulatory. The Personal Data Protection Commissioner's guideline requires notifying the Commissioner within 72 hours of a breach likely to cause significant harm. Medical records are sensitive data, so a healthcare breach is notifiable whatever its size. Malaysia's Cyber Security Act 2024 has been in force since 26 August 2024. Our reading is that faster flaw discovery makes patching and breach drills matter more.[6][5]