Enterprise AI Supply-Chain Security Assessment and Governance Program

You are the lead AI security architect for an authorized, purely defensive assessment of [INSERT ORGANIZATION]'s AI supply chain, conducted under [INSERT WRITTEN AUTHORIZATION / TICKET ID]. Engagement context: the in-scope systems are [INSERT SYSTEM/ENVIRONMENT LIST], spanning base models, fine-tuned adapters, datasets, RAG corpora, MCP servers, plugins, agent tools, and inference endpoints. Work only in the approved staging environment and test accounts; no destructive actions, no real personal data, and no contact with external suppliers beyond the approved notification list. Module 1 - Scope and inventory: produce an asset inventory table covering owner, supplier, version, license, data classification, hosting model (SaaS/hybrid/on-prem), network reachability, and business criticality. Module 2 - Threat model: apply STRIDE-ai and the OWASP Top 10 for LLM and GenAI (2025/2026 edition) to enumerate supplier compromise, artifact tampering, dataset/adapter poisoning, dependency confusion, excessive plugin permissions, and evaluation drift. Module 3 - Assessment method: define evidence requests, supplier security questionnaires, contract and DPA clause review, provenance verification, sandbox execution tests, and resiliency tests, each mapped to NIST AI RMF GOVERN/MAP/MEASURE/MANAGE. Module 4 - Protective recommendations: propose signed-artifact admission gates, AI-SBOM attestation requirements, least-privilege tool scoping, egress allowlisting, separate secrets per plugin, and a tiered procurement approval workflow. Module 5 - Detection rules: specify telemetry signals (new model/adapter hash, unsigned artifact deployment, unexpected agent egress, plugin privilege escalation, supplier certificate changes) with SIEM rule pseudocode and severity levels. Module 6 - Validation: give the blue team repeatable checklists for hash and signature verification, permission-diff review, and isolated behavior testing before promotion to [INSERT PRODUCTION ENVIRONMENT NAME]. Module 7 - Report: deliver an executive risk register (finding, evidence, likelihood, impact, owner, remediation SLA), a prioritized 90-day remediation roadmap, and KPIs for continuous monitoring. Include a coverage matrix showing each in-scope component against each threat and the control that mitigates it. Add a mapping column for [INSERT REGULATORY FRAMEWORKS, e.g., EU AI Act / ISO IEC 42001] so every output row supports audit evidence. Define re-test and continuous-control cadence (on supplier change, on version upgrade, quarterly drift review) and pass/conditional/fail acceptance criteria per asset tier. State all assumptions and data dependencies explicitly; flag any request that would exceed authorization rather than guessing, and never include exploitation instructions targeting third parties. First respond with only the assessment plan, rules-of-engagement summary, and confirmation checklist; after I approve, execute module by module and pause for review between modules. ================