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

Reference · November 2026

Pentesting glossary: the words this site uses

This pentesting glossary defines the model and testing words used on bestaiforpentesting.com, in the sense the linked sources use them.

Each entry is a definition for an authorized test, not a procedure.

Updated November 18, 2026Authorized work only
01A to Z

Definitions

Abliteration
An edit that reduces the refusal direction in an open-weight model, not a fine-tune and not permission.
Authorized testing
A test of a system you own, a signed scope, a bug bounty within its rules, a CTF or a lab.
Base URL
The host an agent sends requests to. Silk Compute uses https://api.silkcompute.com, and /v1 is added only when the tool asks.
Black-box testing
A test with no source, from output you already collected, inside the scope.
Chat Completions
The OpenAI-style API at POST /chat/completions. Silk Compute also serves Messages and Responses.
Context window
How much text one request can hold. Every Silk Compute model here is 1,048,576 tokens.
CTF
A capture-the-flag event or a lab machine you were given. Ask for an approach, not the flag.
CVSS
The Common Vulnerability Scoring System, from FIRST, used here as the categories the notes support rather than an invented score.
MCP
Model Context Protocol, so an agent calls a tool server such as HexStrike AI while the model stays the agent's model.
Messages API
The Anthropic-style API at POST /v1/messages, which Claude Code speaks and Silk Compute serves.
OpenAI-compatible
A server that accepts the same JSON shape as OpenAI's chat API.
Penetration test
An authorized assessment that looks for weaknesses and reports them, following the PTES and NIST SP 800-115.
Reasoning effort
A setting of low, high or max, with max for attack-path analysis, and a small max output on a long-thinking model returns nothing.
Red team
An authorized operation with an objective, allowed tactics and a debrief, which is not permission to add tactics.
Refusal
A model or classifier declining a request, which policies require for unauthorized intrusion and which also stops some allowed work.
Rules of engagement
The hours, rate, emergency contact and data limits in the scope, which a model does not get to loosen.
Scope
The assets, accounts, dates and exclusions in the statement of work, and anything else is not a target.
White-box testing
A test with the source and the running system, which is what Shannon is built for, and the source does not extend the scope.
PTES
The Penetration Testing Execution Standard. It names the stages from pre-engagement through reporting.
OWASP WSTG
The OWASP Web Security Testing Guide, a checklist for an authorized web test. A category is not a recipe.