Key takeaways
- OMB Memorandum M-26-04 requires new federal LLM solicitations to request an acceptable use policy, model or system cards, end-user resources and a feedback channel.
- An October 20 DoD memo gives contracting offices standard solicitation language and makes this documentation a formal part of the proposal process.
- The VA's Enterprise AI Support Services buy, estimated at $25 million to $90 million, lists transparent AI governance among its goals.
- Write down your data handling, testing, human oversight and portability plans now, and share them with buyers before the RFP drops.
What changed this fall
For years, many AI vendors treated governance as something to sort out after award. Federal buyers are moving it earlier. Three recent documents explain why.
- OMB Memorandum M-25-22 (April 3, 2025) sets the rules for how civilian agencies buy AI. It covers contracts awarded under solicitations issued from late September 2025 on, plus options exercised after that date. It tells agencies to test proposed solutions before selecting one where practicable, to avoid vendor lock-in, and to bar vendors from training their models on nonpublic agency data without consent.
- OMB Memorandum M-26-04 (December 11, 2025) implements Executive Order 14319 and its two "Unbiased AI Principles": truth-seeking and ideological neutrality. New solicitations for large language models must include contract terms on those principles, and agencies had until March 11, 2026 to update their procedures.
- A Defense Department memo dated October 20, 2026, from John Tenaglia, Principal Director of Defense Pricing, Contracting, and Acquisition Policy, applies the same direction to Pentagon AI buys. DefenseScoop reports that it gives contracting offices standardized solicitation language and makes the vendor's compliance documentation a formal part of the proposal process.
Put together, these mean that for many AI buys, the documents describing how your product works and how it is governed are now something you submit with your bid.
What buyers are asking for
M-26-04 is the most specific. It says solicitations for LLMs must ask vendors for:
- An acceptable use policy, usually written by the model's developer.
- Model, system or data cards covering training summaries, known risks and mitigations, and benchmark scores.
- End-user resources such as tutorials and developer guides.
- A way for users to report outputs that violate the principles. A general inbox or product contact can satisfy this.
Agencies may also ask for more, such as bias evaluation results or details on changes third parties made to a model. The memo also tells agencies to modify existing LLM contracts where practicable, no later than the next option exercise. If you already have an AI contract with an option coming up, expect this conversation.
The DoD memo adds a practical wrinkle. How much information a vendor can provide depends on where it sits in the supply chain. A reseller or integrator has less direct access to a model's training details than the company that built it. If you're in the middle of that chain, get the documents from your model provider now, not the week a proposal is due.
The VA buy to watch
The Department of Veterans Affairs is a good example of where this is heading. In September it posted a sources sought notice (36C10B26Q0834) for Enterprise Artificial Intelligence Support Services, with an estimated value of $25 million to $90 million. Nextgov reported that the contract would run three years, likely firm-fixed-price, to support a goal of 540,000 provisioned users rolled out in six waves.
The VA lists "run transparent AI governance" among the goals of its broader AI effort, and the notice says the work should help "ensure effective governance of AI initiatives." The RFI also asked industry how the VA could run a technical challenge as part of the evaluation. The VA said it planned to release a final solicitation in October, so check SAM.gov for the current status before you plan around it.
What to have ready before the RFP
None of this requires a new department. It requires writing down what you already do, in a form a contracting officer can read. A short checklist:
- Collect the core documents. Acceptable use policy, model or system card, end-user guides and a named feedback contact. If you resell or integrate, get these from your model provider in writing.
- Document your data handling. Where customer data goes, how long you keep it, and a plain statement that agency data isn't used to train models without consent.
- Describe testing and monitoring. What you test before release, what you measure after deployment, and how you handle a problem report.
- Show human oversight. Who can review, override or switch off the system, and how that's logged.
- Plan for portability. M-25-22 asks agencies to avoid lock-in. Be ready to explain how a customer could move its data and models if it left.
Our guide to selling AI to government covers the broader buying process. If you also sell to states, our state AI policy tracker shows where state rules are going.
Use it early, not just in the proposal
Governance documents are also a good reason to talk with buyers before a solicitation drops. Program and acquisition staff are working out what to ask for, and many would welcome a vendor who shows up with clear answers instead of a pitch.
A respectful first note might read: "We've put together our model card and data-handling summary in the format the new OMB guidance describes. If it would help as you plan upcoming AI requirements, I'm happy to send it over. If the timing isn't right, no problem." It gives the buyer something useful and leaves them room to say no.
Sources sought notices, RFIs and new agency guidance are all early signals worth tracking. Our guide to government buying signals explains how to watch for them.
How KSG helps
Kensington Sales Group's market intelligence team tracks AI requirements, sources sought notices and policy changes across federal and state agencies. Our SDRs then use those signals to open low-pressure conversations with the program and acquisition staff who are shaping the next solicitation.
Sources: OMB M-26-04, Increasing Public Trust in Artificial Intelligence Through Unbiased AI Principles (December 11, 2025); OMB M-25-22, Driving Efficient Acquisition of Artificial Intelligence in Government (April 3, 2025); DefenseScoop, Pentagon issues guidance on AI procurements (October 23, 2026); Nextgov, Veterans Affairs previews timeline for enterprise AI services competition (September 23, 2026); HigherGov, VA Enterprise Artificial Intelligence Support Services, 36C10B26Q0834; iQuasar, Week 1 Nov 2026: GovCon Key Developments (November 3, 2026).