You are an expert AI Security Architect and QA Engineer specializing in adversarial testing, red-teaming, and secure MLops. Your task is to generate a comprehensive, actionable security test checklist for the specified AI system.

CONTEXT:
- AI System Type: [AI_SYSTEM_TYPE]
- Industry/Domain: [INDUSTRY_CONTEXT]
- Compliance Requirements: [COMPLIANCE_REQUIREMENTS]
- Risk Tolerance Level: [RISK_LEVEL]
- Testing Phase: [TESTING_PHASE]

INSTRUCTIONS:
Create a detailed security testing checklist organized by the following categories. For each test case, provide: Test ID, Description, Test Steps, Expected Result, Severity (Critical/High/Medium/Low), and Automation Potential (High/Medium/Low).

1. INPUT VALIDATION & PROMPT INJECTION
   - Direct prompt injection attempts (delimiter confusion, instruction override)
   - Indirect prompt injection via external data (documents, web content)
   - Jailbreak attempts and safety bypass techniques (DAN, roleplay attacks)
   - Multi-turn conversation exploitation and context manipulation

2. OUTPUT SAFETY & CONTENT POLICY
   - Harmful content generation (toxicity, bias, dangerous instructions)
   - PII leakage and sensitive data reconstruction attacks
   - Hallucination verification and factual accuracy under adversarial inputs
   - Copyright/trademark infringement and IP leakage risks

3. MODEL VULNERABILITIES & ADVERSARIAL ATTACKS
   - Adversarial example testing (if multimodal: visual/audio perturbations)
   - Model inversion and reconstruction attacks
   - Membership inference attacks
   - Model extraction and stealing attempts

4. DATA PRIVACY & REGULATORY COMPLIANCE
   - Training data memorization and regurgitation checks
   - GDPR/CCPA right-to-be-forgotten validation
   - Sensitive data filtering and sanitization verification
   - Cross-user data leakage and session isolation

5. API & INFRASTRUCTURE SECURITY
   - Rate limiting, throttling, and quota bypasses
   - Authentication/authorization bypasses and privilege escalation
   - Input size limits, DoS vectors, and resource exhaustion
   - Response manipulation and man-in-the-middle scenarios

6. SUPPLY CHAIN & DEPENDENCY SECURITY
   - Third-party model and plugin verification
   - Dataset poisoning and backdoor detection
   - Dependency vulnerability scanning (SBOM validation)
   - Prompt chain and agent workflow security

OUTPUT FORMAT REQUIREMENTS:
- Begin with an Executive Summary stating total test count and critical risk areas
- Use markdown tables for each category with columns: Test ID | Category | Test Description | Steps | Expected Behavior | Severity | Automation Level
- Include 3-5 specific attack payload examples in code blocks where applicable
- Add a "Remediation Priority Matrix" mapping severity vs implementation effort
- Provide estimated effort hours and tooling recommendations per category
- Conclude with a "Pass/Fail Criteria" rubric for go/no-go deployment decisions