← Capture Corner

Which File Your Product Lands In

A generative AI capability that enters VA as a commercial license lands in a three-column file with no risk fields. The same capability entering as a use case carries seven attestations and an independent review by an office outside development. That routing decision is made at requirement definition, before you ever see the solicitation.

Capture Corner VA AI Governance Risk: Post-Award Reclassification

★ Premium Capture Corner

VA runs two AI inventories and they carry different obligations. One holds 367 use case records, each with seven risk management attestations behind it. The other holds product entries with three columns and no risk fields at all, and that is where Microsoft Copilot Chat sits, inside a $4.65 billion enterprise agreement nobody bid against. This Capture Corner works the buy: the classification decision that sets your compliance cost before a requirement is written, the ambient scribe portfolio and the enterprise IDIQ about to consolidate it, the two Chief AI Officer instruments that closed in June with no award reported, the four flowdown roles and government benchmarking rights in proposed GSAR 552.239-7001, and why selling AI into DHA runs on a different motion entirely. Premium members read the full brief. Subscribe at missionmeetstech.com/pricing.

See premium plans

Capture Corner is the premium BD intelligence companion to Mission Meets Tech. Public-record sourced. Independent analysis. Not a recommendation, not vendor advocacy, not capture material. Built for federal health BD, capture, and proposal leaders who need analytical depth, not headlines.

This issue is the companion to "Filed Under Document Drafting," published this week. The public piece reads VA's inventory files against the Inspector General's findings. This one works the buy: the classification decision that sets your compliance cost before a requirement is written, the money already sitting under the Copilot license band, the ambient scribe portfolio and the enterprise IDIQ that will consolidate it, the two Chief AI Officer RFIs that closed this summer and what the follow-on has to contain, the four flowdown roles and government benchmarking rights in proposed GSAR 552.239-7001, and the asymmetry between selling AI into VA and selling it into DHA.


1. The picture at a glance

Field Detail
Triggering document VA OIG 26-00182-140, published June 11, 2026 [CC1]
VA individual AI use case inventory 367 records, 215 high-impact, 134 deployed, 90 deployed and high-impact [CC2]
Consolidated commercial inventory structure Three columns: use case, product, license band. No risk fields. [CC3]
Microsoft 365 Copilot Chat Absent from all 367 individual records; appears only in the commercial file at the 50,000+ band [CC2][CC3]
VA Microsoft enterprise agreement $4,650,679,410.16, Dell Federal Systems L.P., order 36C10B25F0093 under GSA MAS 47QTCA22D003G, April 2025 to March 2030 [CC4]
VA GPT unit economics $1.25 per user per month; about 100,000 users onboarded [CC5][CC6]
Ambient scribe awards Abridge 36C10G25C0007 at $5,369,500; Knowtex 36C10G25C0013 at $5,452,500 [CC7]
Enterprise consolidation vehicle Ambient Scribe Enterprise IDIQ 36C10B26R0006, TAC New Jersey; GAO protest B-424447.1 (Vexcel) dismissed May 19, 2026 [CC8]
Governance instrument in market RFI 36C10B26Q0537, Enterprise AI Governance, Management, and Implementation Support, responses closed June 23, 2026 [CC9]
Interface and API instrument Sources sought 36C10B26Q0485, VA Enterprise Artificial Intelligence, responses closed June 9, 2026 [CC10]
Proposed acquisition clause GSAR 552.239-7001, 91 FR 36559, published June 17, 2026, comments closed August 3, 2026 [CC11]
DoD position Exempt from M-25-21 inventory and high-impact posting under Section 3(b)(v) [CC12]

Confidence: High on every inventory figure, computed directly from VA's three published spreadsheets. High on the OIG findings, read from the final report. High on the GSAR and OMB text, read from the Federal Register and the memoranda. Medium on the contract values, which come from trade and aggregator coverage of award records rather than from the contract documents themselves; verify in the federal award record before pricing against them. Medium on the competitive analysis in Sections 4 through 7, which is my construction from public solicitation descriptions.

The lead finding for a capture team: at VA, a generative AI capability entering as a commercial license lands in a three-column file with no risk fields, and the same capability entering as a use case record lands in a pipeline with seven attestations, an impact assessment, and an independent review by an office outside development. That routing decision is made at requirement definition, before you see the solicitation, and it moves your compliance cost by a factor a bid team can actually calculate.


2. The classification mechanic, stated precisely

Confidence: High. Computed from the published files.

VA maintains two separate AI inventories and they carry different obligations.

The individual use case inventory holds 367 records. Each carries owning office, development stage, high-impact status, topic area, development type, and, for deployed high-impact cases, seven minimum risk management practice attestations: pre-deployment testing, AI impact assessment, independent review by an office not involved in development, ongoing monitoring, periodic operator training, fail-safe, and appeal process. [CC2]

The consolidated commercial inventory holds product entries with three fields. Use case description, product name, estimated license band. [CC3]

Microsoft 365 Copilot Chat is in the second file, grouped with Power Automate AI Builder and Microsoft Dynamics D365 CoPilot under "Generating first drafts of documents, briefing, or communication materials using AI," at the 50,000-plus band. A full-text search of all 367 individual records returns no Copilot Chat entry. [CC2][CC3]

Why this is a capture fact and not a policy curiosity. The commercial inventory has no high-impact field, which means a product recorded there cannot be designated high-impact, which means the seven attestations never attach. A competitor whose functionally similar capability is procured as a use case rather than as an enterprise license carries pre-deployment testing, an impact assessment, an independent review, and a monitoring plan as delivery cost. Same software category, different bid.

Where the decision gets made. Not by the vendor. It follows from how the requirement is written and which office buys it, and neither is visible to industry until the solicitation posts. The practical move is to ask the question during market research, in writing, and to price both paths until you have an answer.

The exposure to name in your risk register. A product sold into the commercial path today can be reclassified later. The OIG's second recommendation asks VHA to determine whether the chat tools warrant high-impact treatment. [CC1] If reclassification happens after award, the attestation package becomes a scope question on a contract priced without it.


3. The money already committed

Confidence: Medium. Award values from trade and aggregator coverage of the award records.

Instrument Value Detail
VA Microsoft ELA, order 36C10B25F0093 under IDV 47QTCA22D003G $4,650,679,410.16 total, $465,772,965.92 action obligation Dell Federal Systems L.P. as reseller, VA Technology Acquisition Center, 2 bidders, NAICS 511210, April 1 2025 to March 31 2030 [CC4]
Prior VA Microsoft ELA $1,590,224,178 Dell Federal Systems, April 2022 to March 2025. Lower confidence, surfaced from a contract record aggregator [CC4]
GSA OneGov Microsoft agreement Announced September 2, 2025 Microsoft 365 Copilot at no cost up to 12 months for Microsoft 365 G5 customers, opt-in through September 2026 [CC13]
VA GPT $1.25 per user per month in pilot No dedicated contract, task order, or vehicle for VA GPT was located; inventory records it as developed with both contracting and in-house resources [CC2][CC5]
Abridge ambient scribe $5,369,500 Contract 36C10G25C0007 [CC7]
Knowtex ambient scribe $5,452,500 Contract 36C10G25C0013 [CC7]
Rise8 and Thoughtworks Not disclosed Ambient scribe rollout and post-MVP development support, awarded March 2, 2026 [CC7]

Three reads for a pipeline.

First, the Copilot exposure is already paid for. A 50,000-plus license band sitting inside a $4.65 billion enterprise agreement, with GSA OneGov terms that made Copilot free for up to twelve months for G5 customers, is the reason this capability spread without a procurement event anybody could bid against. There is no displacement play here. The play is adjacent: governance, assurance, monitoring, and the interface layer.

Second, the ambient scribe portfolio is pre-consolidation and multi-vendor. VA's inventory carries at least six distinct ambient documentation efforts across OIT and VHA with different vendors, all at pilot or pre-deployment stage, all with blank attestation fields because VA's two-step process completes at step one for cases that have not deployed. [CC2] That fragmentation is what an enterprise IDIQ exists to end.

Third, the individual scribe contracts are small. Five million dollars each is a pilot-scale buy for a system with 172 medical centers. The enterprise vehicle is where the real dollars go, and the protest is already resolved.


4. The instruments in market and what closed

Confidence: High on the solicitation records. Medium on the forward read.

Two Chief AI Officer instruments closed this summer and neither has a publicly reported award as of this analysis date.

RFI 36C10B26Q0537, Enterprise Artificial Intelligence Governance, Management, and Implementation Support. Responses closed June 23, 2026, twelve days after the OIG report published. The stated scope is a multidisciplinary contractor team covering the AI use case review lifecycle, regulatory compliance tracking, governance products, and an enterprise AI landscape assessment. [CC9]

Read that scope against the OIG's three recommendations and the overlap is close to complete. VA is buying the capability to execute the remediation it committed to in its May 19, 2026 response. This is the single most direct procurement consequence of the report, and a market research response closing that fast after publication usually means the requirement predated it and the report accelerated it.

Sources sought 36C10B26Q0485, VA Enterprise Artificial Intelligence. Published May 29, 2026, responses closed June 9, 2026. VA sought a new front-end AI interface plus API access for the workforce, with LLM API services currently running through VA Enterprise Cloud across AWS GovCloud, Azure Government, and Azure Commercial. VA disclosed 95,000 VA GPT sign-ups and more than 5,500 daily users inside the RFI itself. [CC10]

The absence of a dedicated VA GPT contract is the notable feature. A tool with 100,000 onboarded users, built with a mix of contracting and in-house resources, co-developed with Microsoft, has no located award instrument of its own. That implies no entrenched incumbent on the interface and API layer, and it is why the sources sought exists.

Also open or recently closed and worth tracking: the Ambient Scribe Enterprise IDIQ 36C10B26R0006 out of TAC New Jersey, issued March 20, 2026 with April responses and a GAO protest by Vexcel Corporation dismissed May 19, 2026. [CC8] And sources sought 36C10B26Q0234, Summit Data Platform Analytics Solution, a PACT Act requirement posted March 10, 2026 to be issued to GSA schedule holders, with an extended response period. [CC14] The Summit prime is not confirmed.

What to watch for. A follow-on solicitation to 36C10B26Q0537 is the event. If it posts as a standalone AI governance services requirement, it establishes VA's first dedicated AI assurance buy, and no comparable standalone model-validation or clinical-AI-validation procurement was located anywhere in the federal health space. First-mover past performance on that scope has an obvious downstream value at DHA, HHS, and IHS.


5. The compliance bar you will actually be measured against

Confidence: High. Read from the Federal Register text and the OMB memoranda.

Three instruments now set the vendor-side bar, and two of them are new enough that most capability statements do not address them.

Proposed GSAR clause 552.239-7001, "Basic Safeguarding of Data Within Large Language Model Artificial Intelligence Systems," 91 FR 36559, published June 17, 2026, listening session July 14, comments closed August 3, 2026. [CC11] GSA's Federal Acquisition Service has separately posted proposed Government AI system terms instructing contracting officers to insert the clause in solicitations for AI capabilities.

Applicability is narrower than it first reads. The clause applies only when Government Data will be processed by an LLM, and expressly does not apply where the LLM is embedded in a common commercial product such as a word processor, or where LLM functionality is incidental to the core requirement. [CC11] That carve-out is worth reading twice alongside Section 2 of this brief, because the same reasoning that files a chat tool under document drafting would also argue the clause does not attach.

Four flowdown roles, each mapped to NIST AI RMF 1.0 Appendix A actor categories: LLM Developer, LLM System Operator, LLM System Integrator, LLM Service Provider. [CC11] Your teaming structure now determines which set of obligations you inherit, which means teaming decisions have a compliance cost that did not exist a year ago.

The substantive obligations that change bid math:

  • Government retains full ownership of all Government Data and Custom Developments, with only a limited revocable license to the contractor.
  • Training, fine-tuning, or otherwise improving an LLM on Government Data is prohibited, including models operated by third parties.
  • Restricted human access to Government Data, automated ingestion without human content review, encryption, and audit logging that does not display the data.
  • Data localization within FedRAMP-authorized services and logical segregation from non-Government customer data.
  • Certified permanent deletion at contract end.
  • Automated detection and user notification to prevent or reject entry of PII into the LLM, where tools are available.
  • Export in open or standard machine-readable formats with documented APIs, no proprietary formats, and migration tooling.
  • The Government reserves the right to run its own automated benchmarks against the production model at any time, and the contractor must provide the tools and interfaces to enable it.
  • Government may suspend use pending remediation, and the contractor is liable for reasonable decommissioning costs on termination for cause after written notice.

Price the last three. Government-run benchmarking against a production model requires an interface most commercial products do not expose. Certified deletion and open-format export are engineering work. Decommissioning liability is a contingent cost that belongs in your risk reserve rather than in your fee.

OMB M-26-04, December 11, 2025, implementing EO 14319, applies to any LLM procured by an agency. Solicitations and orders issued after December 11, 2025 must include compliance terms, and agency policies including user reporting processes were due March 11, 2026. Minimum vendor disclosures are an Acceptable Use Policy, Model, System, and Data Cards, End User Resources, and a mechanism for end user feedback. [CC15] Those four artifacts are cheap to produce and expensive to be missing at proposal time. If you do not have them written, that is a one-week task.

The Joint Commission's Responsible Use of AI in Healthcare certification, launched June 1, 2026, is voluntary and certifies organizations rather than products, across five standard areas, and does not require Joint Commission accreditation to pursue. CHAI released governance playbooks across eight domains on May 27, 2026, framed as a route to it. [CC16]

The VA-specific angle on that certification: VHA is Joint Commission accredited by directive under VHA Directive 1100.16, with 176 VA programs accredited in 2025. [CC17] The OIG cites Joint Commission and CHAI guidance directly in its report. That gives the certification's elements accreditation-adjacent weight at VA rather than purely advisory weight, and it is the nearest thing to an organization-level AI attestation currently on the market.


6. Selling into DHA is a different motion

Confidence: High on the exemption text. Medium on the market implication.

DoD is exempt from M-25-21's AI strategy requirement under Section 2(a) footnote 9 and from the annual AI use case inventory and public posting requirement, including high-impact designation, under Section 3(b)(v). [CC12]

The practical consequences for a capture team working both agencies:

There is no public DHA AI inventory to mine. At VA you can read 367 records, see which office owns what, see development stage, and build a displacement map from a spreadsheet. No equivalent artifact exists on the DoD side, so DHA market research runs on solicitations, industry days, and program office conversations rather than on a published portfolio.

The designation gate does not exist, which cuts both ways. DHA's Clinical AI Agent inside MHS GENESIS and its enterprise ambient listening deployment carry no high-impact designation process. [CC18] A vendor does not have to price the seven attestations. A vendor also cannot use them as a discriminator, because the customer has no field to score them in.

DHA has generated the evidence anyway. The fall 2024 DHA and CDAO generative AI red-teaming exercise ran with more than 200 clinician and analyst participants across DHA, the Uniformed Services University, and the services, producing more than 800 bias and vulnerability findings on clinical note summarization and medical advisory chatbot use cases. [CC1] Structured adversarial evaluation of a clinical language model is a demonstrable capability with a named DoD precedent. If you can do it, DHA already knows what it looks like.

The JPSR seam is a joint problem with a joint budget. JPSR is co-owned and co-operated by DHA and VHA and funded through the Joint Incentive Fund. [CC19] VA's National Center for Patient Safety asked DHA in 2025 to add an AI cause code and no decision was reached. [CC1] A capability that instruments AI-attributable safety events has two customers, one funding mechanism, and an open unresolved request sitting in the record. That is an unusual configuration and it favors whoever brings a proposal to both sides at once.


7. The counterargument, stated fairly

Confidence: Medium. This is analysis, not a document finding.

Three reasons a governance-heavy capture thesis could underperform.

The evidence base is thin. No documented incident anywhere attributes a clinical documentation error specifically to a general-purpose chatbot. The OIG's own JPSR search returned zero AI-related reports across roughly 180,000 annual submissions. [CC1] Requirements that get written on prospective risk move slower than requirements written after an event, and a pipeline built on the first can sit unconverted for years.

The regime you would be selling compliance into has never recorded a failure. Across all 90 deployed high-impact use cases, the April 2026 attestation columns show no "No" responses on any of the seven minimum practices. [CC2] If the attestations function as a checkbox in practice, the assurance market attached to them is smaller than the policy stack suggests.

And the direction of travel is toward acceleration. M-25-21 itself directs removal of bureaucratic requirements that inhibit innovation, EO 14179 revoked the prior AI executive order, and VA has kept shipping. VA ChatEHR was announced August 4, 2026, nearly two months after the report, having handled more than 53,000 questions from over 8,000 users since April. [CC20] VA is also expanding agentic AI through a $1.6 billion Salesforce enterprise agreement. [CC21] Adoption velocity has not slowed while the governance backlog grows.

The counter to the counter: VHA's own action plan commits to penalties for inappropriate use of generative AI tools. [CC1] That is a governance program with an enforcement component, and enforcement components require monitoring infrastructure somebody has to build.


8. What to do in the next ninety days

Ask the classification question in writing. On every federal health AI pursuit, ask the contracting officer during market research whether the capability will be recorded as an individual AI use case or as a commercial product entry, and whether high-impact designation is anticipated. Price both paths until you get an answer. This costs one email and it changes your delivery cost estimate materially.

Produce the four M-26-04 artifacts this month. Acceptable Use Policy, Model and System and Data Cards, End User Resources, and an end user feedback mechanism. Any LLM you sell to any agency needs them, solicitations after December 11, 2025 must include compliance terms, and having them written is a proposal-day advantage over competitors who will scramble.

Price the GSAR clause obligations before it finalizes. Government-run benchmarking against your production model, certified deletion, open-format export with documented APIs, and decommissioning liability on termination for cause. Comments closed August 3, so the next event is the final rule. Engineering lead time on benchmark interfaces is longer than proposal lead time.

Map your teaming to the four flowdown roles. Developer, Operator, Integrator, Service Provider. Decide which role each team member holds before you sign a teaming agreement, because the clause assigns obligations by role and a mismatch discovered at award is a renegotiation.

Set an alert on 36C10B26Q0537. A follow-on solicitation to the AI governance RFI is the first dedicated federal health AI assurance buy located in this research. Being on a team for it is worth more than the contract value, because the past performance transfers to DHA, HHS, and IHS.

Build the DHA proposal for the JPSR cause code. The request has been open since 2025, the platform is jointly funded, and no one has a decision. A concrete implementation proposal put to both DHA and VHA National Center for Patient Safety addresses a documented gap with two customers and an existing funding mechanism.

If you sell ambient documentation, position for consolidation rather than for another pilot. Six fragmented efforts, an enterprise IDIQ with a dismissed protest behind it, and a $5 million pilot price point together describe a market about to have one vehicle. Being on that vehicle is the whole game.


Mary

Mission Meets Tech Premium


Next Capture Corner: scheduled for the next major federal health or defense-health opportunity in active capture. Subscribers will receive an alert when the next issue publishes.

Capture Corner is an independent intelligence product. It is not affiliated with the Department of Veterans Affairs, the Defense Health Agency, or any company named here, and it is not connected to any offeror's bid strategy. Premium subscription includes access to the standing intelligence sidebar plus the searchable archive.


Sources

[CC1] Department of Veterans Affairs, Office of Inspector General, "Review of Generative Artificial Intelligence Chat Tools for Clinical Use," Report 26-00182-140, June 11, 2026. Source for the three recommendations and April 2027 targets; the zero AI-related JPSR results against roughly 180,000 annual submissions; VHA's action plan commitments including penalties for inappropriate use; the 2025 National Center for Patient Safety request to DHA for a JPSR AI cause code and the absence of a decision; and the fall 2024 DHA and CDAO generative AI red-teaming exercise participant and findings counts. https://www.vaoig.gov/sites/default/files/reports/2026-06/vaoig-26-00182-140_-_final.pdf

[CC2] Department of Veterans Affairs, "VA AI Use Case Inventory 2025," compliance-updated file published April 2026. Source for the 367 records, 215 high-impact, 134 deployed, 90 deployed and high-impact counts; the seven minimum risk management practice attestation fields and the counts across all 90 deployed high-impact cases; the absence of any Microsoft 365 Copilot Chat record; record VA-24-3086 VA GPT and its development type; and the six ambient documentation records at pilot or pre-deployment stage with blank attestation fields. https://department.va.gov/ai/wp-content/uploads/sites/26/2026/04/VA-AI-Use-Case-Inventory-2025-Web-Compliance-Updates.xlsx

[CC3] Department of Veterans Affairs, "VA AI Consolidated Inventory 2025." Source for the three-column structure and the Microsoft Copilot Chat entry at the 50,000-plus license band grouped under "Generating first drafts of documents, briefing, or communication materials using AI." https://department.va.gov/ai/wp-content/uploads/sites/26/2026/01/VA-AI-Consolidated-Inventory-2025-Web.xlsx

[CC4] OrangeSlices AI, "Veterans Affairs inks a 5-year $4.7B Microsoft Enterprise Agreement," and HigherGov contract record for 47QTCA22D003G-36C10B25F0093. Source for the $4,650,679,410.16 total value, the $465,772,965.92 action obligation, Dell Federal Systems L.P. as awardee, the April 2025 to March 2030 period of performance, VA Technology Acquisition Center as contracting office, two bidders, and NAICS 511210. The prior $1,590,224,178 ELA figure comes from a contract record aggregator and was not independently fetched; treat as lower confidence. Verify all values in the federal award record before pricing. https://orangeslices.ai/veterans-affairs-inks-a-5-year-4-7b-microsoft-enterprise-agreement/

[CC5] Department of Veterans Affairs, "Building the Future: VA's Strategy for Adopting High-Impact Artificial Intelligence." Source for the VA GPT pilot cost of $1.25 per user per month. https://department.va.gov/ai/building-the-future-vas-strategy-for-adopting-high-impact-artificial-intelligence-to-improve-services-for-veterans/

[CC6] Department of Veterans Affairs, "Compliance Plan for OMB Memorandum M-25-21." Source for approximately 100,000 VA GPT users onboarded, the April 3, 2026 deadline for deployed high-impact use cases, the two remedies of waiver or removal from operations, VA's statement that it has issued no waivers, and the accelerated ATO process granting an initial VA authorization within 60 days. https://department.va.gov/ai/department-of-veterans-affairs-compliance-plan-for-omb-memorandum-m-25-21/

[CC7] OrangeSlices AI coverage of VA Digital Health Office ambient scribe awards; government contract award notice for 36C10G25C0013; GlobeNewswire, "Rise8 and Thoughtworks Awarded Ambient Scribe Rollout and Post-MVP Development Support Contract with the U.S. Department of Veterans Affairs," March 2, 2026. Source for the Abridge contract 36C10G25C0007 at $5,369,500, the Knowtex contract 36C10G25C0013 at $5,452,500, and the Rise8 and Thoughtworks award. Award values from trade coverage; verify in the federal award record. https://orangeslices.ai/va-digital-health-office-inks-contract-for-ai-enabled-cots-scribe-solution/

[CC8] SAM.gov notice and HigherGov opportunity record for 36C10B26R0006, "Ambient Scribe Enterprise IDIQ (VA-26-00012979)," Technology Acquisition Center New Jersey; OrangeSlices AI protest record. Source for the March 20, 2026 issuance, April 2026 responses, and GAO protest B-424447.1 by Vexcel Corporation dismissed May 19, 2026. https://sam.gov/workspace/contract/opp/800f1ed1a4584289b1abc8ed2583749d/view

[CC9] SAM.gov, RFI 36C10B26Q0537, "Enterprise Artificial Intelligence Governance, Management, and Implementation Support," VA Chief AI Officer. Source for the June 23, 2026 response deadline and the stated scope covering the AI use case review lifecycle, regulatory compliance tracking, governance products, and enterprise AI landscape assessment. https://sam.gov/workspace/contract/opp/55d40b7c67924675a02091d03f17b386/view

[CC10] SAM.gov, sources sought 36C10B26Q0485, "DA01 VA Enterprise Artificial Intelligence (VA-26-00070193)," Technology Acquisition Center New Jersey, published May 29, 2026, responses due June 9, 2026; FedScoop, "VA seeks information on AI interface, API, workforce," June 1, 2026. Source for the front-end AI interface and API scope, the VA Enterprise Cloud environments named, and the 95,000 VA GPT sign-ups and 5,500-plus daily users disclosed in the RFI. https://sam.gov/workspace/contract/opp/82fb63b32b9e4dbea68f65814491a3e1/view

[CC11] General Services Administration, "General Services Acquisition Regulation; Acquisition of Information and Communication Technology," proposed clause 552.239-7001, 91 FR 36559, June 17, 2026, comments closed August 3, 2026. Source for the applicability language and the commercial-product and incidental-functionality carve-outs; the four flowdown roles mapped to NIST AI RMF 1.0 Appendix A; Government data ownership; the prohibition on training or fine-tuning on Government Data; human access restrictions, encryption, and audit logging; FedRAMP data localization and logical segregation; certified deletion; PII detection and notification; open-format export and migration tooling; unbiased AI principles compliance; the Government right to run its own benchmarks against the production model; suspension pending remediation; and decommissioning cost liability. GSA's Federal Acquisition Service has separately posted proposed Government AI system terms instructing contracting officers to insert the clause. https://www.federalregister.gov/documents/full_text/text/2026/06/17/2026-12205.txt

[CC12] Office of Management and Budget, Memorandum M-25-21, April 3, 2025. Source for the DoD exemption from the AI strategy requirement at Section 2(a) footnote 9 and from the annual inventory and public posting requirement including high-impact designation at Section 3(b)(v). https://www.whitehouse.gov/wp-content/uploads/2025/02/M-25-21-Accelerating-Federal-Use-of-AI-through-Innovation-Governance-and-Public-Trust.pdf

[CC13] General Services Administration, "Multibillion dollar GSA OneGov agreement with Microsoft brings steep discounts," September 2, 2025. Source for Microsoft 365 Copilot at no cost for up to 12 months for Microsoft 365 G5 customers and the opt-in period through September 2026. https://www.gsa.gov/about-gsa/newsroom/news-releases/multibillion-dollar-gsa-onegov-agreement-with-microsoft-brings-steep-discounts-09022025

[CC14] HigherGov opportunity record for 36C10B26Q0234, "Summit Data Platform Analytics Solution (VA-26-00024220)," posted March 10, 2026, to be issued to GSA schedule holders, response period extended. The prime is not confirmed. https://www.highergov.com/contract-opportunity/da01-summit-data-platform-analytics-solution-va-26-00024220-p-04f19/

[CC15] Office of Management and Budget, Memorandum M-26-04, "Increasing Public Trust in Artificial Intelligence Through Unbiased AI Principles," December 11, 2025, implementing EO 14319. Source for applicability to any LLM procured by an agency, the requirement that solicitations and orders after December 11, 2025 include compliance terms, the March 11, 2026 deadline for agency policies including user reporting processes, and the four minimum vendor disclosures. https://www.whitehouse.gov/wp-content/uploads/2025/12/M-26-04-Increasing-Public-Trust-in-Artificial-Intelligence-Through-Unbiased-AI-Principles.pdf

[CC16] The Joint Commission, "Joint Commission releases first-of-its-kind, exclusively designed for healthcare organizations, voluntary Responsible Use of AI in Healthcare certification," June 1, 2026; Coalition for Health AI, governance playbooks release, May 27, 2026. Source for the certification structure, its five standard areas, its organization-level rather than product-level scope, and the absence of an accreditation prerequisite. https://www.globenewswire.com/news-release/2026/06/01/3304442/0/en/joint-commission-releases-first-of-its-kind-exclusively-designed-for-healthcare-organizations-voluntary-responsible-use-of-ai-in-healthcare-certification.html

[CC17] Department of Veterans Affairs, VHA Directive 1100.16, and VA News, "Joint Commission accredits 176 VA programs in 2025." Source for the VHA accreditation requirement through the national Healthcare Accreditation Services contract and the 2025 program count. https://www.va.gov/vhapublications/ViewPublication.asp?pub_ID=9865

[CC18] Defense Health Agency, "Ambient listening to support warfighters," July 6, 2026, and Defense Healthcare Management Systems, "DOD Healthcare Management System Modernization Fact Sheet," as of February 2026, Distribution A. Source for the Clinical AI Agent within MHS GENESIS and for the February 2026 start of full DoD enterprise deployment of Ambient Listening. Vendor and governance details for the Clinical AI Agent are not disclosed. https://dha.mil/News/2026/07/06/13/20/Ambient-listening-to-support-warfighters

[CC19] Defense Health Agency and Veterans Health Administration, "Defense Health Agency, Veterans Health Administration collaboration standardizes patient safety reporting," DVIDS. Source for JPSR joint ownership and Joint Incentive Fund funding. https://www.dvidshub.net/news/472733/defense-health-agency-veterans-health-administration-collaboration-standardizes-patient-safety-reporting

[CC20] MeriTalk, "VA rolls out chatbot to assist staff using the Federal EHR," August 2026. Source for the VA ChatEHR announcement of August 4, 2026, the more than 53,000 questions handled from over 8,000 users since April 2026, and VA's statements that the tool does not access veteran health records and does not assist with clinical decision-making. https://www.meritalk.com/articles/va-rolls-out-chatbot-to-assist-staff-using-the-federal-ehr/

[CC21] MeriTalk, Veterans Affairs coverage. Source for VA's use of a $1.6 billion Salesforce enterprise agreement to expand agentic AI. Trade coverage; verify in the federal award record before pricing. https://www.meritalk.com/news/civilian-agencies/veterans-affairs/

Sources verified as of August 24, 2026.

Want a custom deep-dive on any line in this brief?

Every figure in this brief reads directly from the public record: VA's two published AI inventories, Inspector General report 26-00182-140, the Federal Register text of proposed GSAR 552.239-7001, OMB M-25-21 and M-26-04, and the live solicitation record covering the $4.65 billion Microsoft enterprise agreement, the ambient scribe awards, the enterprise IDIQ, and the two Chief AI Officer instruments that closed in June. Want any one line turned into a one-page position for a specific capability, program office, or vehicle? Reply to this issue and tell me the office and the vehicle.

$50one-time · 5–7 business day turnaround

I will pull the additional public-record sources, read the visible signal, and write a 4–6 page custom intelligence memo on the area you select. You name the question; I do the BD-grade analysis.

Buy a Custom Deep Dive — $50 →

Examples: Deep Dive — a classification path analysis for one named capability across both VA inventories · Deep Dive — a GSAR 552.239-7001 flowdown map priced against a specific teaming structure · Deep Dive — an ambient documentation position built for the enterprise IDIQ rather than another pilot