Google announced on Sept. 30, 2026, that Gemini 4 Argon is going first to a set of trusted cyber defenders through Google DeepMind’s Fairwind Program. For those defenders and Google’s internal teams, Google said it will release the model without cyber guardrails so they can use its full cybersecurity defense capabilities.
Google describes Gemini 4 Argon as a frontier model built for long, complex workflows across software engineering, enterprise knowledge work, and cybersecurity defense. The model is not broadly available yet, and Google has not named a firm date for general access. Google said it is gathering feedback from early testers, taking part in the U.S. government’s voluntary pre-release model access process, and strengthening safeguards before wider release to developers, enterprises, and consumers.
When broader access begins, Google said paid API customers and Google AI Ultra subscribers will be first in line. The company listed introductory API pricing at $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at a 95% discount. After the introductory period, Google said pricing will move to $4 per million input tokens and $20 per million output tokens.
Argon also raises the ceiling on model output. Google said the model expands the output token limit to 1 million tokens from the previous 64,000-token level, giving it more room for long reasoning and multi-step work in a single run.
Why the cyber access stands out
Google said Argon can autonomously find, validate, and patch critical software vulnerabilities. It also said Wiz is already using the model for cybersecurity defense through the Scan for Good initiative, which focuses on protecting public infrastructure by finding and remediating high-risk exposures.
According to Google, Argon uncovered a critical vulnerability exposing sensitive personal information in healthcare software used by hospitals worldwide. Google said previous frontier models had missed the risk. The company did not identify the affected software in its announcement, so the claim should be treated as an early Google-reported result rather than a public vulnerability disclosure with independent technical detail.
Google also cited cyber benchmarks. On CWE-bench v1, which evaluates vulnerability remediation, Argon tied for first with a 68% score. On Google’s internal vulnerability benchmark, the company said Argon found exposures across codebases spanning 20 programming languages. On Wiz’s internal black-box penetration testing benchmark, Google said Argon outperformed Gemini 3.8 Flash Cyber in attack surface discovery, vulnerability identification, and proof-of-concept evidence generation.
The gate around the model
On its Fairwind Program page, Google DeepMind says the program currently works with more than 650 partners globally, but only a set of partners get exclusive access to Gemini 4 Argon. The program prioritizes governments, national cyber authorities, critical infrastructure operators, core technology platforms, and other organizations tied to societal resilience.
Fairwind partners are restricted to defensive and research purposes, including authorized threat simulation, reverse engineering, and malware analysis. Google says malicious tasks such as creating malware are not permitted. Participating organizations must use controls such as user-level authentication, phishing-resistant MFA, and applicable access controls, and partner organizations cannot share, redistribute, or sell access to the frontier models.
What safeguards remain
The no-cyber-guardrails statement applies to trusted defenders and internal Google teams, not to broad release. Google says Argon is designed to refuse harmful cyber and chemical, biological, radiological, and nuclear requests while preserving legitimate dual-use scientific research.
The company also says it is using internal and external red-team testing, prompt-injection training, internal activation monitoring, misalignment monitoring that can stop execution, and hardened sandboxes before wider availability.
Argon is not only a security model
Although the rollout is cyber-first, Google is also positioning Argon as a general frontier model for coding and professional work. The company said Google engineers have been using it for daily debugging, large codebase migrations, algorithm design, and research tasks.
Google said Argon scored 77.9% on DeepSWE v1.1, which measures long-horizon software engineering tasks. The company also said Argon ranked first on AutomationBench with a 51.3% score and led several knowledge-work evaluations, including finance and legal benchmarks.
Google’s evaluation document says non-Gemini comparison results are sourced from model providers’ self-reported numbers unless otherwise stated, and several Argon results are self-computed. That makes the figures useful for early context, but independent testing will matter once access broadens.
What this means for security teams
For security leaders, Argon is less a broadly available product than a signal about where defensive AI is moving. Vulnerability research, exploit validation, patch generation, and black-box testing are becoming targets for long-horizon AI systems that can work across code, tools, and live services.
The tradeoff is that the same capability can support harmful activity if released too broadly. Google’s gated approach may appeal to governments and critical infrastructure operators that want earlier access to defensive technology. It also leaves unresolved questions about how vetting decisions are made, how misuse is detected after access is granted, and how independent reviewers can evaluate privately tested capabilities.
If the approach holds, future frontier AI launches may look less like one public release date and more like a sequence of controlled access tiers.
Get new small business insights by email
Practical ideas and useful articles to help you make better business decisions.
Tech Help Canada Staff researches, writes, and reviews practical content for business owners and professionals. Our coverage spans business, marketing, SEO, technology, and the tools and systems people use to grow and operate online. We focus on clear, useful information backed by research, hands-on experience, and editorial review. Learn more about our team and editorial standards. Need help with something? Contact Us
