Google's Newest AI Has No Safety Guardrails for 650 Groups, and Everyone Else Still Can't Use It
That's an unusual sequencing for a flagship AI launch. Normally the general public, or at least paying developers, get a new model on day one, with enterprise and government access following later through dedicated programs. Argon runs that order backward: the most capable, least-restricted version goes to a few hundred vetted security teams first, while everyone else waits with no committed date.
The reason is specific to what this model is built to do. Argon is tuned for autonomously finding, verifying, and patching security vulnerabilities, a capability that is just as useful for breaking into systems as for defending them.
What the Fairwind Program actually is
Google launched Fairwind on September 2, 2026, as a limited-access program pairing an earlier model, Gemini 3.8 Flash Cyber, with CodeMender, Google's AI agent for vulnerability remediation, per Security Boulevard. Eligible organizations are governments, Google Cloud customers, and cybersecurity partners, with priority given to critical-infrastructure operators and maintainers of widely used software. Participants must restrict access to staff working in cybersecurity, incident response, or penetration testing, and use multi-factor authentication.
Argon became Fairwind's new flagship model on September 30. Trusted defenders inside the program get it with the extra cyber-specific restrictions removed entirely, the "without cyber guardrails" language Google itself used, because those restrictions exist to stop exactly the kind of vulnerability-hunting work Fairwind members are vetted to do.
| Who | Access today | What's confirmed |
|---|---|---|
| Fairwind Program (~650 orgs) | Live now, guardrails removed | Governments, Google Cloud customers, cybersecurity partners who passed vetting and use MFA, per Security Boulevard. |
| Paid API customers, Google AI Ultra subscribers | Not yet. No date given. | Named as next in line, per The New Stack, but Google hasn't committed to a window. |
| Everyone else | No access, no announced plan | Not mentioned in Google's rollout language as of this writing. |
The numbers behind the launch
650+
organizations already inside the Fairwind Program
governments, Google Cloud customers, and cybersecurity partners, per Security Boulevard.
$2 → $4
introductory to standard price per million input tokens
output runs $10 introductory, $20 standard; Google hasn't said when the introductory window ends, per Google's announcement.
0
fixed date Google has given for public or general API access
The pattern behind the restriction
Google has done this before. In July 2026 it introduced Gemini 3.5 Flash Cyber through a similarly limited pilot, citing the dual-use risk of a model that's equally good at finding bugs to fix them and finding bugs to exploit them. Argon extends the same logic to a far more capable model, with VentureBeat describing the release as Google "retaking the benchmark lead over OpenAI and Anthropic, but in limited release," per VentureBeat's coverage.
The New Stack put it more bluntly in its own headline on the launch: Argon "is here, it's great, and you can't have it yet," per The New Stack. That's an accurate summary of where things stand for anyone outside the 650 organizations Google has already vetted.
Can I use Gemini 4 Argon right now?
Only if your organization is a vetted member of Google's Fairwind Program. There is no public access, no general API access, and no confirmed date for either.
What is the Fairwind Program?
A limited-access program Google launched September 2, 2026, giving governments, Google Cloud customers, and cybersecurity partners access to its most capable cyber-defense AI models. Members must restrict access to cybersecurity, incident-response, or penetration-testing staff and use multi-factor authentication.
Why did Google remove safety guardrails for Fairwind users specifically?
Argon's core job, autonomously finding and patching security vulnerabilities, requires doing exactly what the model's cyber-specific guardrails normally block. Google restricts that unrestricted version to vetted defenders rather than lifting it for everyone.
When will Gemini 4 Argon be generally available?
Google has said paid API customers and Google AI Ultra subscribers are next in line after Fairwind, but has given no date for either group or for general public access.
What is Gemini 4 Argon actually good at?
Google positions it for long-horizon coding, legal and financial research, and autonomous cybersecurity work, pairing it with CodeMender to find, verify, and patch vulnerabilities. It ships with a 1 million token context window.
Try the tools ✦
Free browser tools that never upload your files.