VA Artificial Intelligence RFI.pdf

Request for Information

Department of Veterans Affairs Enterprise Artificial Intelligence

VA-26-00070193

May 29, 2026

1. Purpose:

This is a Request for Information (RFI) only issued for conducting market research.

Accordingly, this RFI constitutes neither a Request for Quote (RFQ), Request for Proposal

(RFP), nor a guarantee that one will be issued by the Government in the future; furthermore, it does not commit the Government to contract for any services described herein. The Department of Veterans Affairs (VA) is not, at this time, seeking proposals or quotes, and therefore, will not accept, review, or evaluate unsolicited proposals or quotes received in response hereto. This notice is not to be construed as a commitment on the part of the Government to award a contract, nor does the Government intend to pay for any information submitted because of this request.

The Government does not reimburse respondents for any costs associated with submission of the information being requested or reimburse expenses incurred for responses to this RFI. The information provided may be used by VA in developing its acquisition strategy and requirements documentation. Any information submitted by respondents to this RFI is strictly voluntary;

however, any information received shall become the property of the Government and will not be returned to the respondent. Respondents are responsible for clearly marking any information they consider proprietary or commercially sensitive on a page-by-page basis. VA will use reasonable measures to protect legitimately designated proprietary information, but cannot guarantee confidentiality under applicable law, including the Freedom of Information Act (5 U.S.C. § 552).

Interested parties are responsible for adequately marking proprietary, restricted, or competition sensitive information contained in their response.

2. Background:

VA serves 18.25 million Veterans and their families, administering benefits, delivering clinical care, and providing Veteran-facing digital services at enterprise scale. The Veterans Health

Administration, which operates under VA, is the most extensive healthcare system in the U.S., serving around 9 million Veterans annually through its 1,200 facilities.

As Artificial Intelligence (AI) capabilities advance rapidly, VA is committed to harnessing AI to improve Veterans' lives, delivering faster, higher-quality, and more cost-efficient services, while upholding the highest standards of safety, security, and equity. This RFI shares information on the current and future state of VA's AI requirements and seeks market insights in preparation for a solicitation, in accordance with Federal Acquisition Regulation (FAR) 10.001.

VA’s vision and priorities for artificial intelligence adoption:

VA intends to be an industry leader in adopting AI to improve Veterans’ lives by delivering faster, higher quality, and more cost-efficient services with strong governance and trust.

There are five priorities:

1. Expand AI access across the workforce

2. Build an AI-ready workforce

3. Reimagine high-impact workflows with AI, automation, and agentic capabilities

RFI – VA Enterprise AI

4. Invest in data and infrastructure for AI adoption

5. Run transparent and effective AI governance

This RFI focuses on Priority 1 (expand AI access) and Priority 3 (reimagine high-impact workflows), specifically the user-facing tool and Application Programming Interface (API) services layers that enable both. Priorities 2, 4, and 5 are addressed through complementary VA programs and are out of scope for this acquisition.

On priority 1, expand AI access across the workforce, there have been early wins:

1. VA GPT (a generative AI chat tool co-developed with Microsoft): 95,000 sign-ups to date; a small subset of active users report saving 2–3 hours per week. 70% of survey respondents report improved job satisfaction.

2. Microsoft GitHub Copilot: 2,000 developers signed up to use AI-assisted software development. Survey respondents report saving 8 hours per week per user. 74% report they can focus on work that is more satisfying to them.

3. In clinical care, there was a 22% decrease in mortality associated with implementation of an AI-assisted clinical decision support tool.

4. With AI-powered colonoscopy devices, there was a 21% increase in adenoma detection.

5. More AI-embedded use cases are available publicly through VA’s 2025 AI Inventory interactive dashboard.

3. Current generative AI use:

The federal AI tool landscape has expanded significantly, consistent with the Administration's directive to leverage commercially available solutions to the maximum extent practicable.

Microsoft's 365 Copilot is now available to federal agencies under enterprise licensing, while the

General Services Administration’s (GSA) OneGov initiative and associated pilots has broadened access to generative AI capabilities across many federal agencies. VA is actively piloting tools from multiple Frontier AI Labs to evaluate user experience, performance, and value at enterprise scale.

As of May 2026, VA has concurrent generative AI pilots for user-facing tools through Microsoft and GSA OneGov – Anthropic's Claude for Government, Google’s Gemini for Government, and

OpenAI’s ChatGPT Enterprise FedRAMP. In addition, large language model (LLM) API services are used through cloud providers, as part of VA Enterprise Cloud: AWS Government, Azure Government, and Azure Commercial, consistent with Executive Order (EO) 14271's directive to leverage commercially available solutions to the maximum extent practicable.

Current adoption of generative AI, user-facing tools with chat capabilities:

Products in Table 1have passed VA's Authority to Operate (ATO) requirements for use with

Protected Health Information (PHI) and Personally Identifiable Information (PII).

https://department.va.gov/ai/ai-use-case-inventory/ https://department.va.gov/ai/ai-use-case-inventory/

Table 1: VA Generative AI Adoption Data (as of May 27, 2026)

VA GPT Copilot* M365

Copilot, licensed

Claude for

Gov

ChatGPT

Enterprise

FedRAMP

Go-live 3/12/2025 Copilot Chat:

6/25/2025

Teams with

Copilot:

05/2026

12/19/2025 2/19/2026-

3/3/2026;

4/13/2026

3/25/2026

Delivery In-house VA and

Microsoft, Azure

OpenAI GPT-

4o model

VA contract, Microsoft

VA contract, Microsoft

GSA OneGov

Anthropic

GSA

OneGov, OpenAI

Status Legacy, no new users

Enterprise contract

Pilot Pilot Pilot

Users provisioned for access

110,000 and notified

Copilot Chat, unlicensed and Teams

(with

Copilot):

~540,000, not notified

14,810 and notified

110,000, not notified

110,000, not notified

Users who logged in at least once since go-live

89,000 NA NA 16,072 3,892

Users who sent at least one chat message since go-live

80,000 NA NA NA 2,973

Active users, last 30 days

35,000 74,290 NA 10,871

2,814

Active users, last 7 days

15,000 45,850 NA 6,990

1,568

Daily active user

5,556 20,870 NA 3,062 577

Input tokens, last 30 days

263.1M NA NA 51.9B 5.88B

Output tokens, last

30 days

180.6M NA NA 820.4M 90M

Notes:

1. As of December 2025, VA has ~540,000 employees and contractors/va.gov email holders.

2. Copilot* = Number of users who used Copilot at least once in any of the following

Microsoft 365 app in the given time period. These apps are Microsoft Teams, Microsoft

Outlook, Microsoft Office, and Microsoft Copilot.

3. Gemini for Gov, available from GSA’s OneGov offer, is in pilot use with 20 users, public data only, no VA data.

4. LLM API usage data from VA's cloud providers (AWS GovCloud, Azure Government, Azure Commercial) is not available for inclusion at this time. VA will provide updated usage context if available prior to solicitation release.

5. Where VA is going, three illustrative examples:

VA is building enterprise-grade AI infrastructure to support the full spectrum of AI interaction, from assistive to collaborative to agentic, spanning compute, foundation models, data integration, orchestration, and user-facing surfaces, with governance embedded at every layer.

This architecture prioritizes commercially available solutions consistent with EO 14271 and the

Federal Acquisition Streamlining Act (FASA) and is informed by deployments at leading AI-forward public sector and enterprise organizations in healthcare and financial services, including government digital services agencies and health systems.

For this specific RFI, the focus is:

1. User-facing layer: User-facing surface(s) with chat and agentic capabilities including advanced tooling (connectors, skills, plugins), managed-assistants, and managed-agents.

2. Compute and foundation model layer: The most direct, programmatic access to closed foundation models from multiple frontier labs.

Together, these building blocks will address VA’s knowledge worker, operator, and engineer/quant needs as users move between assistive, collaborative, and agentic based to improve internal operations and Veterans services.

Knowledge Worker: Oncology Clinician

Draft, research, analyze, and execute clinical knowledge

Assistive Collaborative Agentic

AI suggests the next step

In-text clinical support

Chat while caring

Summarize, document, prepare

Delegate the follow-through

Personalized patient education

AI surfaces relevant drug interactions, treatment protocols, or lab trends while the clinician reviews the

EHR. Clinician stays in full control. AI reduces lookup time and cognitive load mid-appointment.

Before appointments, AI synthesizes patient’s oncology history, recent imaging, and care team notes into a pre-visit brief. After, the clinician dictates rough notes. AI drafts a structured

After the appointment, AI reads the visit notes, the patient's treatment plan, literacy level, and preferred language, and then generates personalized take-home education materials covering next steps, warning signs, EHR note for review and sign-off.

and follow-up instructions.

Clinician reviews before sending.

e.g. pull recent lab trends from CPRS/VistA and JLV;

flag contraindications;

Suggest ICD-10 codes e.g. pre-visit prep brief; post-encounter note draft; clinical scribing via ambient recording e.g. " Generate take-home materials for this patient's first chemotherapy cycle in

English at a 6th-grade reading level, consistent with

National Institute of Health

(NIH) plain language guidance

High clinician control Shared: clinician reviews and signs

AI executes, and clinician approves

<-- More clinician control --> More AI autonomy with clinician approval

Operator: Housing Services Manager

Multi-step work across tools, with humans in the loop

Assistive Collaborative Agentic

AI suggests the next word

Draft, summarize, translate policy

Chat across data and documents

Research, analyze, and plan

Delegate the workflow

Connect Veterans to housing resources

Housing manager asks AI to summarize U.S. Department of Housing and Urban

Development (HUD) housing regulations, VA produced internal documents and external internet content, draft policy language, or translate a dense Code of

Federal Regulations (CFR) section into plain language for field staff. AI reduces reading and writing time;

manager makes all decisions.

Manager converses with AI to map local housing resource availability in a target community, identify service gaps against Veteran population data, and model the impact of a proposed policy change. AI is a research and analysis partner.

Agent receives a Veteran referral, checks local housing partner availability, matches the Veteran's eligibility and needs against available programs, drafts an outreach communication, and schedules a follow-up. With each step, flags the manager when judgement is required.

e.g. "Summarize 38 CFR Part

62 as a one-pager”; “Format this memo I wrote in VA-provided template” e.g. "What are the bed availability gaps in the

Phoenix, VA area?"; Gap analysis against VA data e.g. "Process this week's housing referrals, match to available community partners, and flag Veterans without a referral”

High human control Shared: Human interprets and decides

--> AI executes workflow, and human resolves exceptions

<-- More human control --> More AI autonomy with human approval

Engineer and Quant

Make critical decisions Draft, research, analyze code and data; use agents to execute

Assistive Collaborative Agentic

AI suggests the next line

Code completion

Chat while coding

Pair programming

Delegate the entire task

Agentic delegation

AI predicts and autocompletes code as the engineer types. Engineer stays in full control; AI reduces keystrokes, catches syntax errors, and surfaces relevant API patterns in real time.

Engineer converses with AI to design architecture, debug logic, refactor for performance, and interpret data outputs. AI is a thinking partner. AI reasons about the problem, not just the syntax.

Quants use chat to explore datasets and validate analysis.

Engineer defines the goal; AI plans, writes, runs tests, interprets failures, and iterates autonomously across tools and files. Quants delegate full data pipeline builds or analysis runs.

Engineer or quant reviews and approves the outcome.

e.g. AI tool inline suggestions; SQL query completion; Python function scaffolding e.g. "Why is this query slow?" "Refactor this for readability" "Explain what this data pattern means" e.g. "Build and test the claims data pipeline, fix any failures, open a PR when done"

High engineer control Shared control High AI autonomy; human reviews and approves

<-- More engineer control --> More AI autonomy

6. VA’s current thinking on solutioning is below. VA welcomes constructive feedback.

As innovative technology evolves in the next few years, VA needs contractual, pricing, and delivery mechanisms that can quickly, easily adapt to advancements in research and applied AI.

On contracting, VA intends to establish the most direct contractual relationships possible with

Frontier AI Labs, as defined below, to maximize adaptability, speed, and safety, consistent with the Administration's policy to utilize the competitive marketplace and the innovations of private enterprise to the maximum extent practicable.

Frontier AI lab is defined as an organization that:

1. Creates, maintains, and applies proprietary large language models;

2. Demonstrates state-of-the-art performance on recognized, third-party industry benchmarks (e.g., MMLU, HELM, BIG-bench, LMSYS Chatbot Arena;

3. Retains direct control over research, model development, deployment, programmatic access;

4. Applies models to enterprise-grade tools with documented capabilities;

5. Offers commercial, machine readable, programmatic offerings with documented capabilities, including Service Level Agreements (SLA) that ensures high reliability and performance;

6. Conducts research and publishes documentation on detection and prevention of malicious uses of AI; and

7. Publishes documentation alongside release of AI models that inform users about what the model does, where it might fail, and how it has been evaluated.

VA is primarily interested in the most direct contractual relationships with frontier AI labs, as defined above. Responses from resellers, aggregators, or other intermediaries should demonstrate unique and compelling value not available through a direct relationship, for example, proprietary integration capability, specialized compliance infrastructure, or differentiated pricing not otherwise accessible to federal agencies.

NOTE: There is no incumbent contractor currently supporting this effort.

On pricing, VA strongly prefers an exclusively usage-based pricing methodology across all surfaces and contract periods. VA is not interested in seat-based, per-user subscription, or flat-fee pricing structures. This preference reflects VA's need for a flexible pricing model that scales with consumption, accommodates rapid technology evolution, and aligns cost with demonstrated value.

VA intends to seek pooled, volume-based pricing from frontier AI labs that:

1. Encompasses all surfaces, including LLM API services delivered via cloud providers, under a unified pricing structure

2. Provides Most Favored Customer (MFC) pricing, with mandatory benchmarking and price adjustments no less than quarterly; and

3. Includes automatic downward price adjustment mechanisms, triggered no less than quarterly, reflecting demonstrated commercial market price trends for comparable AI services.

On surfaces, VA’s focus for this RFI is in two areas:

1. User interface tool(s):

a. Tool(s) from one Frontier AI Lab that address the user types, workflows, and needs described above in assistive, collaborative, and agentic capabilities with AI. Such capabilities span from knowledge work, data analysis, code generation, and automation. Enterprise work apps with AI embedded (e.g. email, calendar, etc.) and domain-specific apps with AI embedded are out of scope (e.g. ambient scribe).

b. Environment that addresses VA Directive 6500, federal laws (Congress), federal regulations (federal administrative agencies authorized by Congress to implement laws) that relate to protecting VA information and information systems through risk management and compliance with federal security laws. This includes workflows where Protected Health Information (PHI) and Personally Identifiable Information

(PII) may be used.

2. API services:

a. The most direct machine-readable, programmatic access, e.g., RESTful API, to AI models, managed-assistants, and managed-agents created by U.S.-based organizations that specialize in developing the most advanced large language models and foundational AI. These organizations are defined by their innovation in research capabilities, substantial computational resources, and focus on pushing the boundaries of machine learning and natural language processing.

b. VA has direct relationships with multiple, proprietary model creators.

c. Access to full feature set on frontier AI labs’ release cadence.

d. Environment that addresses VA Directive 6500, federal laws (Congress), federal regulations (federal administrative agencies authorized by Congress to implement laws) that relate to protecting VA information and information systems through risk management and compliance with federal security laws. This includes workflows where Protected Health Information (PHI) and Personally Identifiable Information

(PII) may be used.

Example of surfaces that cover user-facing tools and API services:

1. First party enterprise tool(s) for knowledge work, data, code, etc.

2. First party API services

3. Third party cloud provider Amazon AWS

4. Third party cloud provider Microsoft Azure

5. Third party cloud provider – other

Out of scope:

1. Enterprise work apps with AI embedded (e.g. email, calendar, etc.)

2. Domain-specific apps with AI embedded (e.g. ambient scribe)

3. LLM API services embedded in existing VA production applications (i.e., API usage already integrated into operational systems, separate from the direct-access API services sought in this RFI)

4. Cloud compute and infrastructure

5. On-device AI (NPU)

6. Self-hosted open-source models

7. Custom model fine-tuning, custom training data

Enablement services:

VA recognizes that software alone is insufficient to apply AI to high-impact internal and external services, at scale, safely and responsibly. VA requires enablement services that build durable internal capacity and accelerate adoption. These services must be commercially available offerings and not custom-developed government-unique solutions. VA is interested in the following:

1. Applied AI or forward-deployed engineers and data scientists embedded within VA operations, with demonstrated coding and data skills, who work directly inside VA environments, including provisioning with va.gov credentials as required

2. AI literacy: Structured products for general and role-specific training

3. AI adoption acceleration: In-app resources, documentation, repeatable playbooks from comparable government, health, or large enterprise deployments

4. AI safety: Tools, documentation, and advisory support to help VA implement responsible

AI governance aligned with VA AI guidance and emerging federal and industry rules and norms

6. Policy and Regulatory Context:

VA's investment in frontier AI capabilities directly advances the Administration's policy of sustaining and enhancing United States global AI dominance. By deploying commercially available AI platforms from U.S.-based Frontier AI Labs at scale across one of the federal government's largest workforces, VA contributes to the national objective of demonstrating that

American AI leadership translates into measurable improvements in public services. Consistent with EO (December 11, 2025), VA seeks to adopt AI applications across its mission areas without imposing government-unique technical requirements that would impede the integration of commercially available innovation.

The following documents are applicable to this requirement:

• EO 14271, Ensuring Commercial, Cost-Effective Solutions in Federal Contracts (April

15, 2025)

• Office of Management and Budget (OMB) M-25-21 Accelerating Federal Use of AI through Innovation, Governance, and Public Trust

• OMB M-25-22 Driving Efficient Acquisition of Artificial Intelligence in Government

• M-26-12, Increasing the Acquisition of Commercial Products and Services (April 17, 2026)

• OMB M-26-04 Increasing Public Trust in Artificial Intelligence through Unbiased AI

Principles

• OMB M-26-10, Reinforcing Transparency, Accountability, and Oversight of Federal

Technology (March 31, 2026)

INFORMATION TO INCLUDE IN A RESPONSE TO THIS NOTICE. Vendors responding to this notice are asked to provide the following:

1. User-facing capabilities:

What user-facing tool(s) does your organization offer that will address where VA is and where

VA will go, on applying AI to improve internal operations and Veteran services?

Consider addressing:

a. Interactive natural language assistants that answer questions, draft and revise content, analyze information through a conversational interface

b. Tool(s) and architecture that enable VA to securely connect VA data sources through standard connections, retrieve, and analyze information, including connection to data in commercial tools such as Microsoft 365 in Microsoft Government Community Cloud

(GCC), VA’s databases in AWS/Azure, and Veteran health data

c. Interactive natural language capability that read, write, test, refactor code and assist with software development

d. Interactive natural language agentic assistant that operates on a user’s desktop or enterprise file repositories to carry out multi-step tasks: organize documents, extract and structure data, draft content, and execute routine workflows with human supervision

e. Multimodal capabilities that include analysis of images, scanned documents, structured forms

f. Enterprise administration capabilities, including usage monitoring, usage enforcement, organizational/workspace-level AI outputs governance, role-based access control, and single sign on

2. API services:

What direct, programmatic service(s) to AI models does your organization offer to address VA needs?

a. Models capable of multi-step reasoning for complex analytical tasks

b. Describe individual, end user experience and capabilities

c. Describe enterprise administration capabilities

d. Compare and contract first party and third-party programmatic access to LLMs. Describe use cases where direct, first-party programmatic access is preferable. Describe use cases where access via a cloud provider is preferable. Describe the key trade-off factors VA should consider when choosing between these access patterns, including factors related to latency, feature availability, pricing, and compliance. Describe scenarios where direct programmatic access from frontier AI labs and programmatic access from cloud providers are important; what would governance model on access, use, evaluation, change look like? Definitions:

1. First-party access: model developer’s own managed service, delivered directly to VA by the organization that develops and operates the model

2. Third-party access: the same or functionally equivalent models delivered through a cloud service provider’s managed AI service.

Requirement/constraint:

1. Third-party offerings must be interoperable with both AWS and Azure environments within VA Enterprise Cloud and other VA cloud service providers.

2. Respondents must also describe the ability and the timeline to extend third-party access, upon VA request, to additional cloud platforms, e.g. Google Cloud and Oracle Cloud.

3. Security:

VA personnel’s use of software to process VA data must be able to obtain a VA ATO and meet policy set by VA Directive 6500 and federal laws and regulations. Describe how your user-facing tool(s) and/or API services currently or will meet this requirement.

4. Enablement service:

What enablement services would you recommend that VA consider? What do you offer directly?

What do you offer with partners?

5. Demand Forecast and Usage Benchmarks:

What is your estimated demand forecast for VA’s use of LLMs?

a. Aggregate, across all surfaces

b. First party, user-facing tool(s)

c. First party, API services

d. Third party, API services

e. Usage benchmarks from other public and private sector organizations of similar size and industry, e.g. government, health, insurance, financial payments, legal, or others.

6. Pricing:

a. On software, what is your usage-based pricing structure for VA?

b. On enablement service, is this included in software pricing? If separate, describe pricing model and provide rough order of magnitude pricing for VA.

c. Describe how you intend to meet VA’s need for quarterly pricing adjustments, if any.

d. If your organization proposes a pooled pricing structure across all surfaces, describe how this structure will be implemented with cloud providers, your organization and partners, and VA.

e. Consistent with OMB M-26-10, Section 2, as part of any future solicitation, VA will request that offerors provide pricing and utilization data in a format that can be shared with OMB and GSA for government-wide benchmarking. Offerors should be prepared to provide pricing information in a machine-readable format consistent with OMB M-25-05.

This information will be used to support government-wide efforts to achieve economies of scale and best-in-class pricing for AI services across the federal enterprise.

f. Are you prepared to provide pricing and utilization data to VA in a machine-readable format that may be shared with OMB and GSA for government-wide benchmarking, consistent with OMB M-26-10? Describe any constraints or conditions you would require.

g. Has your organization previously provided pricing data to a federal agency for cross-agency sharing? If so, describe the mechanism and any lessons learned.

7. Product(s) delivery

a. How does your organization's AI offering(s) meet the definition of a “commercial product or service” under FAR 2.101? Describe how your products have been deployed for other federal agencies or for non-government customers in the same or similar configuration as what you would propose for VA, without government-unique customization.

8. Additional Information:

Interested Contractors shall also provide the following information in the submission:

• Name of Company:

• Address:

• Point of Contact:

• Phone Number:

• Email address:

• Company Business Size and Status:

o For VOSB and SDVOSBs, proof of verification in the Small Business

Administration (SBA) VetCert Database.

o For firms who are not SBA VetCert verified, please provide VOSB and/or

SDVOSB firms you are working with.

• NAICS code(s):

• System for Award Management (SAM) Unique Entity Identifier (UEI):

• Existing Contractual vehicles (General Services Administration (GSA)/ NASA Solutions for Enterprise-Wide Procurement (SEWP), etc.)

• The ability to comply with VAAR 852.219-75 - Notice of Limitations on

Subcontracting—Certificate of Compliance for Services and Construction (Jan 2023)

(Deviation)

9. Response Instructions:

Responses shall be submitted electronically by X1:00 PM Eastern Standard Time, June X09, 2026, via email to the Technology Acquisition Center (TAC) point of contact Contract Specialist

Amber Davis, Amber.Quivers-Davis@va.gov and Contracting Officer, Joe Jones, Joseph.Jones6@va.gov. Please note “RFI# 36C10B26Q0485 – Enterprise

AI36C10B26QXXX” in the subject line of your email response.

Responses shall not exceed 20 pages (excluding cover page and appendices) and shall be submitted in Microsoft Word format, with a minimum 11-point font and 1-inch margins.

Responses exceeding the page limit may not be reviewed in their entirety. VA encourages concise, substantive responses over comprehensive but unfocused ones. NO MARKETING

MATERIALS ARE ALLOWED AS PART OF THIS RFI. Generic capability statements will not be reviewed.

OFFERORS ARE ADVISED THAT THIS REQUIREMENT MAY BE DELAYED,

CANCELED, OR REVISED AT ANY TIME

Formatted: Not Highlight mailto:Kendra.Casebolt@va.gov