AI Governance for Government

Verifiable evidence for Executive Order 14110, OMB AI memoranda, agency AI inventories, and federal contractor AI usage.

Federal AI governance has shifted from voluntary frameworks to enforceable evidence requirements. Executive Order 14110, OMB Memorandum M-24-10, and agency-specific guidance impose documentation, inventory, risk assessment, and impact assessment requirements on federal AI use. H33 produces the cryptographic evidence layer that satisfies these requirements without depending on the vendor that produced the AI.

The federal AI governance landscape

Executive Order 14110 (October 2023) directs federal agencies to manage AI use through risk-based governance, including impact assessments for rights-impacting and safety-impacting AI, requirements for senior AI officials, AI use case inventories, and procurement requirements for AI vendors. OMB Memorandum M-24-10 (March 2024) provides implementation guidance, specifying minimum practices for AI governance, AI use case inventory format, impact assessment requirements, monitoring requirements, and the role of agency Chief AI Officers. OMB Memorandum M-24-18 (October 2024) provides AI procurement guidance, including evidence and documentation requirements for AI products and services. Agency-specific guidance translates OMB into implementation. The common thread: federal AI governance requires per-use-case documentation, per-decision evidence in higher-risk classifications, ongoing monitoring evidence, and procurement-time documentation.

What federal AI governance requires

Federal AI governance requirements at the operational level include: AI use case inventory — every covered AI use must be inventoried with risk classification, ownership, documentation links, monitoring status. Impact assessments — for rights-impacting and safety-impacting AI, impact assessments document intended use, expected effects, mitigation measures, and monitoring approach. Monitoring evidence — ongoing monitoring produces evidence of continued safe and effective operation. Incident documentation — AI-related incidents and near-misses are documented with supporting evidence. Procurement documentation — AI products procured by the federal government must be supported by vendor-provided documentation. Public-facing documentation — certain federal AI uses require public-facing disclosure documents.

How H33 supports federal AI governance requirements

For the use case inventory. Each AI use case has an associated policy. The PolicyBind digest in H33 bundles ties bundles to the inventory entry. For impact assessments. The bundle's PolicyBind ties decisions to the impact assessment version. ModelFingerprint ties the actual model to the assessed model. CorpusBind ties the actual data to the assessed data sources. For monitoring evidence. Continuous monitoring produces aggregate metrics and per-decision evidence. The H33 bundles for each in-scope decision provide cryptographic per-decision evidence. For incident documentation. When an AI-related incident occurs, the H33 bundles for the implicated decisions provide cryptographic evidence of what the system did. For procurement documentation. AI vendors can provide H33 bundles as part of their evidence package.

The federal vendor relationship

Federal AI procurement is increasingly conditional on vendor-provided evidence. Vendors selling AI products to federal agencies face evidence requirements that exceed commercial standards: documentation of training data provenance, evidence of evaluation against agency-specified benchmarks, demonstration of safety and rights protections, ongoing monitoring evidence, incident reporting capability. H33 evidence bundles support each. For vendors competing in federal AI procurement, the H33 evidence layer is increasingly a procurement differentiator.

Use cases

A civilian agency's benefit eligibility AI. The agency uses AI to assist in benefit eligibility determinations. The AI is classified as rights-impacting. The agency's impact assessment specifies the model, the policy, the data sources. The H33 bundles for each eligibility decision reference the impact assessment version. An appeals review or class-action investigation verifies the bundles offline. A DoD intelligence analysis AI. A DoD program uses AI to support intelligence analysis. The bundles for each analysis decision document the data consulted, the model used, the analyst authority. Subsequent IG review or congressional inquiry verifies the bundles offline. A regulatory agency's enforcement AI. A federal regulatory agency uses AI to support enforcement priority-setting. The bundles document the policy, the model, the underlying data. Subsequent legal challenges can verify the bundles to confirm the agency's basis.

Common questions

Does this satisfy EO 14110 requirements?
EO 14110 specifies governance requirements; H33 produces the evidence layer that supports compliance.

What about OMB M-24-10?
H33 bundles satisfy the evidence-related practices (use case documentation, impact assessment evidence, monitoring evidence, incident documentation) and the procurement-side documentation requirements in M-24-18.

Can this be deployed in classified environments?
Yes. The bundle generation and verification can be deployed in classified environments.

Does this work with FedRAMP authorization?
Yes. The bundle-generation system can be deployed on FedRAMP-authorized infrastructure.

How does this map to NIST AI RMF Profile for Federal Use?
As the profile finalizes, H33 EC objects will be mapped to specific profile requirements. The bundle's schema versioning supports evolving alignment.

Get Started

Run the demo Download the verifier Download a bundle