The Pulse
Google Limits Gemini 4 Argon to Vetted Cyber Defenders
Google is giving selected cybersecurity partners access to Gemini 4 Argon through its Fairwind Program before a wider release. The program restricts who can use the model and which security tasks they can perform.
AI.info Team · Written with AI ·
“Safely releasing frontier capabilities at this level requires a phased approach.”
Koray Kavukcuoglu, SVP, Google DeepMind and Chief AI Architect, Google
Google announced Gemini 4 Argon on September 30, 2026, and is giving the model first to selected cybersecurity defenders through its Fairwind Program. The company says the staged release lets early users test the system while Google gathers feedback and strengthens safeguards ahead of wider availability. Developers, enterprises and consumers are next in line, but Google has not given a release date.
Fairwind puts access behind organizational checks
Google says Fairwind works with more than 650 partners globally, but that figure describes the program as a whole, not the number of organizations receiving Argon. The company prioritizes governments and organizations responsible for critical infrastructure, including healthcare, telecommunications, energy and financial services, as well as technology platforms whose software supports many other users. Applicants undergo background checks focused on security history and ethical operations.
Partner access comes with rules about who may use the model and for what. Organizations must limit access to their internal cybersecurity, incident-response or penetration-testing teams, use user-level authentication and phishing-resistant multifactor authentication, and track employee use. Approved work includes authorized threat simulation, reverse engineering and malware analysis for defensive or academic research; creating malware is prohibited, and partners cannot sell or redistribute access.
Google says Argon found a flaw in hospital software
Google says Argon can autonomously find, validate and patch critical software vulnerabilities. The company also says it will provide trusted defenders and Google’s internal teams with Argon “without cyber guardrails,” so those users can test its full cybersecurity capabilities. That model-level change does not remove Fairwind’s partner rules: the program still limits users, permitted work and distribution.
Google points to an early test with Wiz, which is using Argon through its Scan for Good initiative. The company says the model found a critical vulnerability in healthcare software used by hospitals worldwide that exposed sensitive personal information and had escaped earlier frontier models. Google’s announcement does not name the software or publish technical details about the vulnerability, so its account does not allow outside readers to assess the finding independently.
Benchmarks and pricing frame the wider rollout
Google reports that Argon scored 68% on CWE-bench v1, tying for first in a test of vulnerability remediation. On the Fairwind page, Google also compares Argon with Gemini 3.8 Flash Cyber: 85.8% to 71.0% on its real-world vulnerability-discovery benchmark, and 70.9% to 58.2% on Wiz’s penetration-testing benchmark. Those are company-published results; the page does not provide enough detail to treat them as independent verification.
The planned model has a maximum output of 1 million tokens, up from 64,000, and Google lists introductory API prices of $2 per million input tokens and $10 per million output tokens. After the introductory period, the announced rates rise to $4 and $20, respectively. Those prices do not make Argon generally available now: Google says broader access will start with paid API customers and Google AI Ultra subscribers, without specifying when.
CodeMender offers another route for defenders
Fairwind partners can use Argon on its own or alongside CodeMender, Google’s code-security agent for automating software fixes. For organizations that do not qualify for the program, Google points to CodeMender with publicly available models and its other AI security products. The distinction is central to the rollout: Argon’s strongest cyber capabilities go first to vetted teams, while the broader release remains pending.