Nine mitigations — built around three principles: verify what you load, limit what it can do, and watch for what you missed
MIT01
Provenance verification and SBOMs/AIBOMs
Verify
MIT02
Dependency gatekeeping — allowlists, pinning, typosquat scanning
Verify
MIT03
Containment and sandboxed builds
Limit
MIT04
Secure prompts and versioned memory
Verify
MIT05
Inter-agent security with mTLS and PKI
Verify
MIT06
Continuous validation and monitoring
Watch
MIT07
Content hash pinning
Verify
MIT08
Supply chain kill switch
Limit
MIT09
Zero-trust design
Limit
The three-layer strategy
Verify (MIT01, 02, 04, 05, 07): Don't rely on name or signature alone — verify integrity, provenance, and content before loading anything. Limit (MIT03, 08, 09): Assume something will slip through — sandbox its execution, restrict what it can access, and plan for rapid isolation. Watch (MIT06): Continuously validate that what's running matches what was approved, and alert when it doesn't.