Answer first: Gemini 3.5 Flash Cyber is not a normal public Gemini model. Google says it will be available exclusively to governments and trusted partners through a limited CodeMender pilot. Do not plan a product around a presumed API key, public model selector, or Flowith entitlement.
What Google actually announced
Gemini 3.5 Flash Cyber is a specialized model built on Gemini 3.5 Flash and fine-tuned for finding and fixing cybersecurity vulnerabilities. Google pairs it with CodeMender, an agent system that coordinates multiple Flash Cyber agents and combines their work into one report.
The deployment boundary matters as much as the model:
| Question | Source-backed answer |
|---|---|
| Is it generally available in the Gemini API? | Google does not announce general API access. |
| Who can access it? | Governments and trusted partners in a limited pilot. |
| How is it delivered? | Through CodeMender, not as a standalone general-purpose chatbot. |
| What is the intended use? | Defensive vulnerability detection, validation, and patching. |
| Is Flowith access confirmed? | No. Check the live Flowith workspace rather than inferring access. |
Google explains that the controlled rollout responds to the dual-use nature of cyber capability. That is an authorization boundary, not a temporary inconvenience to route around.
What a serious evaluation would require
An eligible organization should evaluate the complete system, not only whether the model finds a bug:
- Define the repositories, languages, vulnerability classes, and environments in scope.
- Isolate the test environment and prevent access to production credentials or sensitive services.
- Require reproducible evidence for every finding.
- Validate patches with tests, static analysis, dependency checks, and human review.
- Measure false positives, missed vulnerabilities, unsafe patches, regressions, review time, and rollback success.
- Record tool calls, changed files, approvals, and the final disposition.
A plausible patch is not a safe patch. It must preserve behavior, close the vulnerability, avoid introducing a new weakness, and pass the organization’s release controls.
Do not blur three different Gemini products
- Gemini 3.6 Flash is a broadly available workhorse for coding, knowledge work, multimodal analysis, and agentic workflows.
- Gemini 3.5 Flash-Lite targets high-throughput, latency-sensitive work.
- Gemini 3.5 Flash Cyber is a restricted cyber model paired with CodeMender.
Shared “Flash” branding does not imply shared access, safeguards, pricing, or intended use.
What to do without access
Do not wait for a restricted model to improve software security. Strengthen the controls already available:
- Keep dependency inventories and automated update policies current.
- Run static, dependency, secret, and infrastructure-as-code checks.
- Use fuzzing and targeted security tests where the risk warrants them.
- Require isolated reproduction and a test that fails before each patch.
- Give any coding assistant the least privilege needed for the task.
- Review generated code and retain a reversible commit trail.
- Use coordinated disclosure and incident processes for real findings.
Generally available AI can assist with explanation, test generation, and patch review where policy permits, but it does not replace authorization, validation, or a security owner.
FAQ
Can I access Gemini 3.5 Flash Cyber through the normal Gemini API?
Google’s launch does not describe general Gemini API access. It says the model will be exclusively available to governments and trusted partners through a limited CodeMender pilot.
What does CodeMender do with Flash Cyber?
Google describes CodeMender as a code-security agent that coordinates multiple Flash Cyber agents to detect, validate, and patch vulnerabilities and produce one combined report.
Is Flash Cyber a general coding model?
No. Google presents it as a specialized cyber-focused model for finding and fixing vulnerabilities inside a controlled agent system.
What should teams use without pilot access?
Keep existing secure-development controls: dependency and static analysis, code review, fuzzing, patch validation, least-privilege test environments, and current generally available models where policy permits.
Official sources
Sources checked July 29, 2026: