Skip to main content
WireRead
Back to all news

Google

Google launches Gemini 3.8 Flash Cyber under vetted Fairwind Program

The security-tuned model pairs with Google's CodeMender harness but is restricted to approved government and industry partners.

By , Editor-in-Chief · WireReadVerified September 2026

The answer

Google released Gemini 3.8 Flash and a gated cybersecurity variant on 2 September.

Google made Gemini 3.8 Flash generally available on 2 September 2026 and launched a security-focused version, Gemini 3.8 Flash Cyber, restricted to a new vetted access scheme called the Fairwind Program.

Google said the programme pairs the model with its CodeMender agent harness "to help defenders find, verify, and fix vulnerabilities at agentic scale".

Gemini 3.8 Flash Cyber is a post-trained version of the standard 3.8 Flash model, built for cybersecurity work and succeeding 3.5 Flash Cyber, according to Google. It shares the same underlying intelligence as the base model but carries fewer restrictions on advanced security tasks, allowing it to perform work the standard model would otherwise block.

The model can search code for vulnerabilities, verify findings, generate fixes and validate patches before deployment, Google said.

Access to Fairwind is by application only, with no self-serve option. Google said eligible participants include government agencies, Google Cloud customers and security partners, with priority given to critical-infrastructure operators and maintainers of widely used software. More than 650 organisations already take part, according to Google, and members must limit access to security staff and apply protections such as multi-factor authentication. Google Cloud customers can still use the CodeMender harness with public models without joining the programme.

In internal tests, Google said the model found over 70% of vulnerabilities across 20 programming languages and produced 2.6 times more correct patches than competing models. On a Collinear patching benchmark, it scored 47.2% pass@1 against 47.8% for a leading frontier model, at a much lower cost, according to Google.

The launch follows similar moves by rivals to restrict their strongest cybersecurity models to vetted users. Anthropic gated its Claude Mythos 5 model behind Project Glasswing in April, and OpenAI limited GPT-5.6-Cyber to its Daybreak programme, according to Google's announcement.

Further detail on the Fairwind Program's rollout was reported by Security Boulevard.

Sources

← All news