Field manual · AI for authorized security testing · Edition November 2026 · Updated 2026-11-18

Phase · November 2026

AI for pentest tooling: the best model, and how to use it

The best AI for pentest tooling is Kimi K3.1 Pro, because the job is parsers, glue and read-only checks, which is coding with tool calls.

DeepSeek V4.1 Flash writes a small parser for much less when the file is already on disk.

Updated November 18, 2026Authorized work only
01Method

Tools that read evidence, and stop there

A parser for nmap XML, a joiner for two CSV exports, or a filter that prints hosts inside the scope. The script reads a file and writes a table. It does not open a network connection. Semgrep rules belong here when they only match source. Claude Code, Codex or OpenCode is the agent.

Rules of engagement

Only on systems you own, a signed scope, a CTF or a lab. A model that answers is not permission.

02With a model

Describe the file, the columns, and the prohibition

Name the file, the fields, and that the program must not open a socket. A short parser is a DeepSeek V4.1 Flash job. Read the diff before you run it.

Prompt

Write a read-only parser for the nmap XML file in this folder. Print hosts, open ports and service names. It must not open a network connection or invoke another program. Comment the fields it expects.

Agent setup is Claude Code, Codex and OpenCode. Output feeds internal network or reporting. Other grades are on the home matrix.