Uncensored AI for security tooling
The best uncensored AI for security tooling is Kimi K3.1 Pro in a coding agent such as Claude Code or Codex. Use it for detection rules, code scanners, fuzzing harnesses and audit scripts on systems you own or are authorized to test.
Work that hosted assistants often decline
- Detection rules for a known threat
- A fuzzing harness for a parser you maintain
- An inventory of services on your own network
All of it is legitimate on systems you own or may work on, as why coding agents refuse explains.
Which model
| Model | Use it for | Price per 1M, in / out |
|---|---|---|
| Kimi K3.1 Pro | Tools that span many files; anything you will ship | $3.00 / $8.00 |
| DeepSeek V4.1 Flash | Rules, parsers and scripts in bulk | $0.24 / $0.90 |
| GLM-5.3 Flash | Many small rules at once | $0.24 / $1.00 |
Silk Compute prices, read November 18, 2026.
Workflow
- Write the scope down
List the hosts, repositories and rules in a file such as
scope.txt, and tell the agent inCLAUDE.mdorAGENTS.mdnever to touch anything outside it. - Give it the context
Let the agent read the code, configs or logs the tool will work on.
- Build in small pieces, with tests
Ask for one function and one test at a time.
- Run it in a lab first
Use a container or a VM, and read every line before it touches anything real.
Common tools are YARA and Sigma for detection rules, Semgrep rules for code patterns, AFL++ for fuzzing your own code, and Nmap's scripting engine for inventory on networks you run.
A prompt to start from
Read scope.txt. Write a Python script that checks every host in scope for TLS certificates that expire within 30 days and for TLS versions below 1.2, and writes the results to a CSV. Add tests against a local test server. Never connect to a host that is not in scope.txt.Paste it into Claude Code, Codex or OpenCode once it runs on an abliterated model.