How to Map NIST AI 300-1 Documentation Standards to SWT3 Witness Evidence

1. Quick Reference

2. What is NIST AI 300-1?

NIST AI 300-1, Guidance and Templates for Public-Facing AI Documentation, is part of NIST's AI Standards Zero Drafts pilot project -- an initiative designed to accelerate AI standards development and broaden participation beyond traditional standards bodies. The pilot produces initial public drafts that invite early feedback from industry, academia, civil society, and government, enabling faster convergence on practical standards for AI governance.

The standard provides practical templates for standardized documentation of AI models and datasets intended for public disclosure. It uses a three-tier field designation system: Required (mandatory for all public-facing documentation), Recommended (organizations should populate when feasible), and Optional (discretionary based on context). This tiering recognizes that not all documentation fields carry equal weight -- some are essential for public trust, while others serve specialized audiences. The standard distinguishes between public-facing documentation (intended for transparency and external stakeholders) and internal technical documentation (intended for governance, audit, and operational teams).

AI 300-1 calls for both machine-readable and human-readable documentation artifacts, reflecting the reality that documentation serves multiple consumers: regulators who read narrative summaries, automated compliance tools that parse structured data, and downstream deployers who need programmatic access to model metadata. The standard explicitly allows security carve-outs: algorithm types, exact model sizes in parameters, and specific hyperparameter configurations may be omitted from public documentation to prevent adversarial exploitation. This carve-out directly aligns with SWT3 clearing levels, which attest to the existence of documentation without requiring content disclosure. AI 300-1 extends the NIST AI RMF "Map" function with operational documentation templates that translate governance intent into concrete, verifiable artifacts.

3. Model Documentation Mapping

Each row maps a NIST AI 300-1 model documentation field to the SWT3 procedures that produce cryptographic evidence satisfying that field's requirements.

4. Dataset Documentation Mapping

Dataset documentation fields from AI 300-1 mapped to SWT3 procedures that produce evidence for data governance and provenance requirements.

5. Control Detail Cards

6. Security Carve-Outs and Clearing Levels

NIST AI 300-1 explicitly allows omitting certain documentation fields from public disclosure: algorithm types, exact model sizes in parameters, and specific hyperparameter configurations. The rationale is straightforward -- full disclosure of these technical details enables adversarial exploitation. An attacker who knows the exact architecture, parameter count, and training hyperparameters can craft more effective attacks against the model. The U.S. Chamber of Commerce specifically recommended these omissions in their formal feedback on the draft, reinforcing that security carve-outs are an expected and defensible practice.

SWT3 clearing levels directly address this tension between transparency and security. Level 0 (Analytics) provides full transparency -- all factor values are visible in the witness anchor, suitable for non-sensitive models where public documentation includes complete technical details. Level 1 (Standard) hashes factor values so that the presence of documentation is attested without disclosing content, suitable for most AI 300-1 "Required" fields where the organization wants to prove documentation exists. Level 2 (Sensitive) preserves the factor structure but redacts values, suitable for hyperparameter documentation where the organization needs to document the existence of configuration choices without adversarial exposure. Level 3 (Classified) provides existence-only attestation, suitable for classified model documentation in national security contexts.

This means organizations can satisfy AI 300-1's documentation requirements at different clearing levels depending on the sensitivity of each field. The witness anchor proves the documentation exists and was verified, even when the content is not public. A model card that says "hyperparameter configuration: documented" backed by a Level 2 AI-MDL.7 anchor is compliant with both the documentation requirement and the security carve-out -- the assessor can verify the anchor without the organization disclosing adversarial-sensitive parameters.

7. Dual Compliance: AI 300-1 + EU AI Act Annex IV

Organizations operating in both US and EU jurisdictions can use a single SWT3 evidence chain to satisfy both AI 300-1 documentation requirements and Annex IV technical file obligations. The procedures listed below produce witness anchors that are jurisdiction-neutral -- the same cryptographic evidence satisfies both frameworks.

8. Quick Reference: Examiner Questions

9. SDK Quick Start

Python

Full SDK documentation: sovereign.tenova.io/docs

Create a free account: sovereign.tenova.io/signup

10. References

• NIST AI 300-1 (Initial Public Draft): Guidance and Templates for Public-Facing AI Documentation
• NIST AI Standards Zero Drafts Pilot Project
• NIST AI Risk Management Framework
• EU AI Act Annex IV: Technical Documentation
• Related: Annex IV Metrics Guide
• Related: Annex VII Data Room Guide
• Related: SWT3 Protocol Guide
• Related: NIST AI 600-1 GenAI Profile Crosswalk
• SWT3 UCT Registry -- 280 procedures across 9 namespaces
• SDK Documentation