Model provenance is the traceable history of a model’s origin, training or calibration data, code, dependencies, parameters, validation, approvals, versions, deployment, monitoring, and retirement. In a digital twin, it connects a physical or operational decision to the exact model state that influenced it.
Control the whole artifact
A filename or checksum is not sufficient when behavior also depends on runtime libraries, feature logic, parameters, configuration, input schema, units, and environment.
Record fitness for purpose
Approval should name the intended operating range, decision context, known limitations, validation evidence, owner, and expiration or review conditions.
Make rollback trustworthy
Preserve known-good artifacts, dependencies, configuration, and data contracts. Test rollback before an incident.
Related resources
References
- NIST IR 8356: Security and Trust Considerations for Digital Twin TechnologyNIST · 2025 · Government research publication
Primary NIST treatment of digital-twin characteristics, cybersecurity, and trust considerations.
- Artificial Intelligence Risk Management Framework (AI RMF 1.0)NIST · 2023 · Framework
Voluntary framework for managing risks associated with AI systems.
- NIST SP 800-160 Vol. 1 Rev. 1: Engineering Trustworthy Secure SystemsNIST · 2022 · Systems-security engineering guidance
Principles and life-cycle activities for engineering trustworthy secure systems.