# Gemini 3.8 Flash Cyber: what Google's new security AI model means for you

> Google launched a cybersecurity AI model, restricted to vetted partners via its Fairwind Program.

*Google is pairing a restriction-eased AI model with a bug-hunting tool, but only vetted defenders can use it.*

By Behzad Hosseini · SuggestedTech
Canonical: https://suggestedtech.com/news/gemini-3-8-flash-cyber-what-google-s-new-security-ai-model-means-for-you

On 2 September 2026, Google released Gemini 3.8 Flash, its newest fast AI model, and alongside it a specialised version built for cybersecurity defenders. That second model, Gemini 3.8 Flash Cyber, is not available to the public. You can only get it by applying to a new scheme called the Fairwind Program.

## What happened

Google made Gemini 3.8 Flash generally available through its API (the interface developers use to plug a model into their own software), under the name gemini-3.8-flash.

At the same time, it released Gemini 3.8 Flash Cyber, a "post-trained" version, meaning it has been further trained after the main model was built, this time specifically for security work. It follows an earlier model called 3.5 Flash Cyber. The underlying intelligence is the same as the standard model, but Google has loosened some of the restrictions that would normally stop an AI from doing advanced security tasks, such as actively searching for weaknesses in code.

The model can search for vulnerabilities, check whether a flaw is real, write a fix, and confirm the fix works before it goes live. Google pairs it with an existing tool called CodeMender, an AI agent system designed, in Google's words, "to help defenders find, verify, and fix vulnerabilities at agentic scale".

Access runs through the Fairwind Program, and it is by application only. Google says it is prioritising government agencies, its Cloud customers, critical-infrastructure operators and maintainers of widely used software. More than 650 organisations are already taking part, according to [Google](https://blog.google/innovation-and-ai/technology/safety-security/fairwind-program/). Members have to restrict access to their security staff and use protections like multi-factor authentication.

In Google's internal tests, the model found over 70% of vulnerabilities across 20 programming languages and produced 2.6 times more correct fixes than rival models. On one patching benchmark it scored close to a leading frontier model, 47.2% versus 47.8%, at much lower cost, according to [Security Boulevard](https://securityboulevard.com/2026/09/google-launches-fairwind-program/).

## What it means for you

If you're not a security professional or don't work for an organisation in the programme, this model isn't something you'll use directly. Anyone on Google Cloud can still use the CodeMender tool with Google's standard public models, just not with the more capable, fewer-restrictions Cyber version.

If your organisation maintains widely used software or critical infrastructure, this is a tool that could eventually help find and patch flaws in systems you rely on, though access depends on approval.

## How to try it

There's no self-serve option. Access to Gemini 3.8 Flash Cyber requires an application to the Fairwind Program, and Google says it is prioritising government bodies, Cloud customers and security partners, especially those running critical infrastructure or widely used software.

Google is not alone in gating its strongest cybersecurity model this way. Anthropic has done the same with Claude Mythos 5 through its Project Glasswing, live since April, and OpenAI restricts its GPT-5.6-Cyber model through a programme called Daybreak.

## Key takeaways

- Gemini 3.8 Flash Cyber can hunt, verify and fix software bugs with fewer safety limits than normal
- You can't sign up yourself; it's only for governments, Google Cloud customers and security partners
- Over 650 organisations already have access, joining similar gated models from OpenAI and Anthropic

## Sources

- [Google's Fairwind Program: Cyber defense tools for trusted partners](https://blog.google/innovation-and-ai/technology/safety-security/fairwind-program/) — Google, 2026-09-02
- [Google Launches Fairwind Program for Gemini 3.8 Flash Cyber Access](https://securityboulevard.com/2026/09/google-launches-fairwind-program/) — Security Boulevard, 2026-09