AIRed Team Framework GitHub

Contribute

Submit an attack pattern

Fill in the fields below and this page generates a schema-valid content file ready to drop into a pull request. Patterns must be framed as how to test defenses against a technique — read ETHICS and the contribution quality bar first. Field-tested patterns from real engagements are what this library values most.

Next available: ATK-026

Applies to

Defensive framing: describe how to test defenses against this technique, not an offensive recipe. No weaponized payloads.

Comma-separate multiple IDs. At least two of the four frameworks must be mapped.

Before this is submittable

  • Name must be at least 8 characters.
  • Select at least one applicable system type.
  • At least 1 prerequisite.
  • Test approach must be at least 80 characters — describe the defensive test, with authorization assumed.
  • At least 2 expected evidence items.
  • At least 2 detection signals.
  • At least 2 defensive recommendations.
  • At least 2 framework mappings (across OWASP LLM, OWASP Agentic, MITRE ATLAS, NIST AI RMF).
  • At least 2 references.
  • Author attribution is required (and permanent).
{
  "id": "ATK-026",
  "name": "",
  "category": "Prompt manipulation",
  "applies_to": [],
  "prerequisites": [],
  "test_approach": "",
  "expected_evidence": [],
  "detection_signals": [],
  "defensive_recommendations": [],
  "framework_mappings": {},
  "ai_controls_catalog_refs": [],
  "references": [],
  "author": "",
  "created": "2026-07-01",
  "last_reviewed": "2026-07-01"
}
  1. Fork emmanuelgjr/AI-RedTeam-Framework
  2. Add this file at src/content/attacks/ATK-026.json
  3. Run npm run validate locally
  4. Open a PR — the template walks you through the submission checklist

Everything on this page runs in your browser. Nothing you type here is sent anywhere.