| Scenario | MIT01 Provenance/SBOM |
MIT02 Allowlist/Pin |
MIT03 Sandbox |
MIT04 Secure Prompts |
MIT05 mTLS/PKI |
MIT06 Continuous Val. |
MIT07 Hash Pin |
MIT08 Kill Switch |
MIT09 Zero Trust |
|---|---|---|---|---|---|---|---|---|---|
| S1: Amazon Q VS Code — Compromised update | ● | ● | ◐ | – | – | ● | ● | ● | ◐ |
| S2: GitHub MCP tool descriptor poisoning | ◐ | ◐ | ● | – | – | ● | ◐ | ● | ● |
| S3: Postmark MCP impersonator on npm | ● | ● | ◐ | – | ● | ● | ● | ● | ● |
| S4: AgentSmith prompt-hub proxy | ◐ | ● | ● | ● | ● | ● | ● | ● | ● |
| S5: Poisoned nx/debug npm — coding agent | ● | ● | ● | – | – | ● | ● | ● | ◐ |
| S6: Agent-in-the-Middle via agent cards | ◐ | ● | ● | – | ● | ● | ● | ● | ● |
MIT02 (Allowlist/Pin) and MIT08 (Kill Switch) cover all six scenarios — together they form the core of agentic supply chain defense. MIT06 (Continuous Validation) and MIT07 (Hash Pinning) are the next-widest controls. MIT05 (mTLS/PKI) addresses the four scenarios involving agent-to-agent trust. MIT03 (Sandbox) limits blast radius across all runtime execution scenarios.