Google opens cyber defense AI to governments and enterprises through Fairwind program

Google DeepMind has launched a program that gives government agencies and enterprise security teams early access to its most advanced cyber defense AI. It's called the Fairwind Program, and it grants access only to vetted organizations — governments, Google Cloud customers, and cybersecurity partners deemed trustworthy. Google said it's prioritizing organizations that underpin critical infrastructure, such as hospitals and telecom networks.

To break it down: Fairwind is the name of the access program itself, while Gemini 3.8 Flash Cyber is the actual AI model running inside it. CodeMender is the tool that takes the vulnerabilities the model finds and turns them into real code fixes. All three pieces have to work together for defense teams to go from spotting a problem to producing a verified patch in a matter of minutes.

What Fairwind actually combines

Until now, security teams have faced a tradeoff. Massive frontier models are expensive to deploy and hard to run across an entire enterprise codebase, while smaller open-weight models often struggle with complex vulnerability fixes and require teams to build their own tooling and infrastructure from scratch. Fairwind is Google's attempt to close that gap.

At the core of the program is a pairing of the CodeMender harness with Gemini 3.8 Flash Cyber, a model built specifically for cybersecurity work. Google says this combination performs the reasoning needed to write and verify code fixes at a much lower operating cost than traditional frontier models. According to the company, vulnerability fixes that used to take weeks of manual work can now become deployable patches in minutes, all within an organization's own secure cloud environment.

We've covered related developments before.

Who can apply, and how

Fairwind isn't something just anyone can pick up and use right away. Participating organizations have to agree to limit access to their internal cybersecurity, incident response, and penetration testing staff, and to adopt safeguards like multi-factor authentication. Google says it's rolling out access to government and enterprise partners gradually, under these strict operating requirements.

Google Cloud customers outside the Fairwind program aren't shut out entirely. Combined with AI Threat Defense, there's still a path to use CodeMender with other models available on the Gemini Enterprise agent platform. That said, Gemini 3.8 Flash Cyber itself is being made available first to Fairwind participants.

Government agencies or enterprise security teams interested in joining can register their interest through the Fairwind Program application page.

What early partners are saying

These organizations tested Gemini 3.8 Flash Cyber on real vulnerability detection and remediation work during a short trial period and shared their impressions.

Partner	Spokesperson	Key takeaway
Armadin	David Slater, Founder and Principal Architect	Said the model matched larger frontier models' performance at lower cost and latency
CrowdStrike	Daniel Bernard, Chief Business Officer	Called it a step forward for industry-wide vulnerability discovery and remediation speed
Palo Alto Networks	Charlie Sestito, Director, CTO Office	Confirmed the model outperformed comparable models across several cybersecurity tasks
Snowflake	Mayank Upadhyay, CSTO	Ran a two-day trial against its public repositories and got results on par with much larger engines
Wiz	Gal Nagli, Head of Threat Exposure Response	Observed a major performance jump over previous models in handling web vulnerabilities

CrowdStrike's Daniel Bernard said, "Google's Fairwind Program brings the industry together to accelerate vulnerability discovery and remediation."

Google's cybersecurity investment tops $100 million

Fairwind builds on Google's long-running investment in cyber defense. With this new round of funding through Google.org, Google's cumulative cybersecurity-related support worldwide has now crossed $100 million. According to the accompanying 2026 U.S. Cybersecurity Impact Report, $36 million has gone into 35 cybersecurity clinics to date, providing free security support to more than 1,250 organizations, including hospitals, public school districts, and municipal water utilities.

If Fairwind is aimed at large organizations like governments and major enterprises, this clinic funding looks like a separate track meant to extend defensive coverage to smaller institutions that can't afford their own security teams.

Official source video

Editor's take

What's notable about Fairwind is Google's choice not to release this model through an open API, but instead to narrow access to roughly 650 partners. The same reasoning power that makes a model good at defense also makes it good at writing attack code, so Google appears to be buying defenders a head start — putting this capability in their hands first, before it's widely available, so they have time to harden their systems. That's a different path from OpenAI and Anthropic, which have leaned toward releasing coding agents broadly. In cybersecurity specifically, Google has chosen controlled distribution instead. The decision itself says something about how the industry is starting to grapple with the dual-use nature of frontier models.

From a practical standpoint, it's hard for most companies to join Fairwind right away. The program mainly targets government agencies and Google's existing large enterprise and security partners, so smaller security teams are probably better off starting with the general path of pairing CodeMender with publicly available models.

Over the coming weeks, expect the list of participating partners to grow, along with likely expansions of CodeMender features for customers outside the Fairwind program. Google has also gestured toward striking a balance with the open-weight community, which makes how long this controlled-access approach lasts worth watching.