Library › Book 7 › Appendix A: Complete AI Security Testing Checklist
Appendix A: Complete AI Security Testing Checklist
17.1🔒Prompt Injection- [ ] Direct injection test suite with 50+ payloads - [ ] Indirect injection test suite covering documents, HTML, structured data - [ ]…
17.2🔒Jailbreak Testing- [ ] Role-playing attacks (DAN, developer mode, debug mode) - [ ] Hypothetical framing (fiction, education, research) - [ ] Token…
17.3🔒Data Leakage- [ ] PII detection (email, phone, SSN, credit card, API keys) - [ ] System prompt extraction resistance (10+ techniques) - [ ]…
17.4🔒RAG Security- [ ] Retrieval poisoning resistance - [ ] Citation accuracy verification - [ ] Hallucinated citation detection - [ ] Context window…
17.5🔒Output Handling- [ ] HTML output escaping - [ ] SQL parameterization of AI output - [ ] Shell command sanitization - [ ] URL validation - [ ] Markdown…
17.6🔒Access Control- [ ] LLM respects user permission levels - [ ] Tool calls respect authorization boundaries - [ ] High-risk actions require confirmation…
17.7🔒Denial of Service- [ ] Input length validation - [ ] Output token limits enforced - [ ] Request timeouts configured - [ ] Rate limiting per user - [ ] Cost…
17.8🔒Supply Chain- [ ] Model weight checksums verified - [ ] ML dependencies scanned for CVEs - [ ] Model sources restricted to trusted registries - [ ] API…
17.9🔒CI Pipeline- [ ] Pre-commit hooks (secrets, quick SAST) - [ ] SAST with AI-specific rules - [ ] SCA for ML dependencies - [ ] DAST against staging - […
17.10🔒Compliance- [ ] AI features classified by risk level - [ ] Transparency disclosure on AI interfaces - [ ] Human oversight mechanism for high-risk…
17.11🔒Production Monitoring- [ ] Real-time PII scanner on responses - [ ] Anomaly detection for unusual patterns - [ ] Rate limiting and abuse detection - [ ]…