Google Introduces Gemini 4 Argon, Initially Available Only to Selected Cybersecurity Defenders

On September 30, Google introduced Gemini 4 Argon, a new model also designed for defensive cybersecurity tasks. However, regular users, companies and the broader developer community will not get initial access. Google is making it available to selected trusted cybersecurity defenders through the Fairwind program.

The limited deployment is part of the company’s cautious approach to a model whose stated capabilities include finding, verifying and fixing critical vulnerabilities. Google says it is further strengthening safeguards against misuse, prompt injection and unwanted model behavior before any potential broader release.

Gemini 4 Argon in the Fairwind Program

According to Google’s announcement, Gemini 4 Argon is initially available through Fairwind, a program intended for selected cybersecurity defenders. Google has not yet announced when the model might become available to additional developers, businesses or consumers. The date for access through an API or as part of the Google AI Ultra service has also not been confirmed.

The company describes the model as a tool for defensive cybersecurity work. It is intended to autonomously assist with finding and validating flaws, as well as designing or carrying out fixes for critical vulnerabilities. According to its own statement, Google is providing it to trusted defenders without cybersecurity guardrails.

This mode is also one of the areas where public technical details are still lacking. Google says access remains controlled and that safeguards against misuse will be further developed before broader deployment. However, the announcement does not specify the rules of use or mechanisms for external oversight of the model’s work without these restrictions.

Price and Output Limit

Google has set the output limit at 1 million tokens. The introductory price is $2 per million input tokens and $10 per million output tokens. The company did not say how long the introductory pricing would remain in effect or what the final price would be afterward.

The million-token output limit suggests use in larger workflows, such as code analysis or longer security tasks. However, the specific practical parameters of deployment will depend on the conditions of the Fairwind program, which Google did not detail in the information provided.

Google Reports Finding a Bug in Healthcare Software

Google, together with Wiz, says the model helped find a critical flaw in healthcare software. However, sufficient technical information has not been published to independently assess the scope of the vulnerability, the vendor affected or the repair status.

It has also not been confirmed that any specific vulnerability described in connection with the announcement was exploited in real-world attacks. Google presents Argon as a preventive defense tool, not as a response to a confirmed security incident.

What Will Matter as Access Expands

Google presents the model’s results and performance on security tasks, but complete independent validation of these measurements was not documented in the available materials. Independent testing therefore remains important when evaluating Gemini 4 Argon, particularly in finding, verifying and fixing vulnerabilities.

Other points to watch will include technical information about the reported healthcare software flaw, any coordinated disclosure and details about the fix. Google will also need to provide more information about security measures, external red-team testing and the rules under which selected defenders will use the model.

The announcement therefore represents a controlled introduction of a model with advanced cyber capabilities, not general availability. Google has not yet publicly decided the scope of further deployment.

Verified and updated: 10/01/2026 06:22