G.1.4 Deliver provenance
Control Details
Objective
Enable traceability and reproducibility of the actions performed while writing code, compiling, testing, and deploying software by making transparent the steps that were performed, by whom and in what order.
Definition
Automatically collect and provide traceability data for generated products, including AI models, including: what entity built the package, what build process they used, and the top-level inputs to the build, such that the data's traceability can be verified and is generated automatically from the build service using a convention common in the package ecosystem. Provenance SHOULD accompany the artifact at publish time, and package ecosystems SHOULD provide a way to map a given artifact to its corresponding attestations.
Assessment Questions
- Describe your manual and/or automated processes for generating provenance data.
- What authentications take place relative to the production of provenance data?
- How is provenance data verified?
- How is the integrity and immutability of provenance data managed?
- What do provenance records include, e.g. the initial state of the machine, VM, or container, and user-specified build steps?
- How do you ensure that access to generated AI models trained on sensitive data (e.g. PII, IP, financial, or other regulated data) is restricted to individuals who already have access rights to the sensitive data or that generated models do not contain sensitive data?
Reference sources
- EO 4e(vi) 4e(vii) 4e(x)
- SSDF PS.3.2
- SSDF-AI PS.3.2
- SLSA Distributing provenance Source L3: Signed and auditable provenance
- 800-161 SR-4
- OWASP-SCVS 6.1 6.2 6.3 6.4
- CNCF-SSC A-A Distribute in-toto metadata with TUF D-V Ensure clients can perform Verification of Artifacts and associated metadata D-V Ensure clients can verify the freshness of files D-A Use The Update Framework
- Self-attestation 3
- SAMM I-SB-2-B
- OSPS OSPS-BR-02 OSPS-BR-04 OSPS-DO-06 OSPS-QA-02