Policy series · Article 4 of 8
Modernizing Government Technology Procurement for AI-Readable Communications
Most agencies publish through platforms they contract for. If AI-readable publishing becomes a policy goal, procurement is where it would be written into practice.
Why procurement matters
Few agencies build their own publishing systems. They license or contract for website platforms, content management, emergency notification and social tools. What those platforms can export, and in what form, largely determines what an agency can offer to AI systems. A standard that agencies cannot implement through the tools they already buy will not be used.
What recent OMB procurement guidance says
OMB M-25-22 (April 3, 2025) addresses how federal agencies acquire AI systems and services. According to a legal summary of the memo, it encourages contracts that avoid vendor lock-in through knowledge transfer, data and model portability and transparency; calls for ongoing monitoring of AI performance; says contracts should bar vendors from using government data to train AI without the agency’s explicit consent; suggests solicitation terms that require vendors to disclose unanticipated AI use; and directs GSA to build a repository of responsible AI procurement resources.
Those principles concern buying AI. A separate question is whether similar expectations should apply when agencies buy the platforms they publish through, since those platforms now sit between the agency and AI systems.
What procurement language could ask for
- Structured export. The agency can export its public statements in an open, machine-readable format with authority, jurisdiction, source and time as named fields.
- Portability. The agency owns its records and can publish them through additional channels, including other registries or feeds, without restriction.
- Verification support. The platform can attach or pass through a hash and signature, and supports withdrawal notices.
- Key custody stated clearly. The contract or policy says who holds the signing keys, how public keys are published, and how keys are rotated and retired. Whichever model is used, the arrangement is documented.
- Disclosure. The vendor discloses where AI is used in processing or transforming the agency’s published content.
Who holds the signing keys is a policy choice
This article discusses policy options and does not prescribe the architecture of any implementation. Signing keys are a good example, because more than one workable model exists.
| Model | How it works | Main trade-off |
|---|---|---|
| Agency-held keys | The agency signs each record with a key it controls. | Strongest agency control, but the agency carries key management, rotation and recovery. |
| Registry-held signing | A registry verifies that a submission comes from the named authority, then signs the record on the agency’s behalf. | Less operational work for the agency, but it relies on the registry’s identity checks and key security. |
| Delegated or shared | The agency and a registry or platform divide responsibilities, for example the agency approves and the registry signs. | Needs clear, documented responsibilities. |
Procurement language can stay neutral among these. What it should require is the outcome: public keys are published, retired keys remain listed so earlier records can still be verified, the party that holds each key is identified, and anyone can check a record without relying on the channel that delivered it.
Policy options
| Option | What it would do | Main trade-off |
|---|---|---|
| Model contract language | GSA and state procurement offices publish sample clauses agencies can adopt. | Voluntary, so adoption depends on each buyer. |
| Requirement in new solicitations | Structured export and portability become expected terms for new publishing platforms. | Vendors must build to it; smaller vendors may struggle. |
| Interoperability profile | Procurement points to a shared profile for fields and verification. | Depends on that profile existing first. |
| Address at renewal | Agencies negotiate terms as existing contracts come up. | Slow, and uneven across agencies. |
Benefits
- Agencies avoid being tied to one vendor’s format for their own public statements.
- Expectations are consistent across buyers, which helps vendors and agencies alike.
- The change arrives through ordinary contract cycles instead of a separate project.
Implementation burden
Existing contracts run for years, so change arrives gradually. Vendors would need development work to support structured export and signing. Procurement staff would need template language and guidance. State and local governments often buy through cooperative or statewide contracts, so those vehicles are the natural place to start.
Drawbacks and open questions
- Box-checking. A clause can be met on paper without improving the result. Requirements should be testable: can a third party verify an exported record?
- Naming technology. Procurement language that names a product, method or key-custody model ages badly. It should specify outcomes and open formats, and stay neutral among providers.
- Competition. The goal is portability. Terms should not favor any one registry, feed or vendor, including those that might later qualify as a standard.
- Reach. OMB memoranda direct federal agencies. State and local procurement would need its own vehicles.
What to resolve first
A shared definition of the minimum fields and how they are verified, so that procurement language has something concrete to point to. See the discussion of provenance standards in this series.
Sources
- Hunton Andrews Kurth, OMB issues revised policies on AI use and procurement by federal agencies (summary of M-25-21 and M-25-22)
- OMB Memorandum M-25-21, Accelerating Federal Use of AI through Innovation, Governance, and Public Trust (April 3, 2025)
- OMB Memorandum M-23-22, Delivering a Digital-First Public Experience (September 22, 2023)
- Foundations for Evidence-Based Policymaking Act of 2018, Public Law 115-435 (Title II, OPEN Government Data Act)
Sources were read in the original or, where noted, through a published summary. Statuses and dates are as of October 18, 2026.