The Pentagon Put ChatGPT and Grok on GenAI.mil. OpenAI and xAI Hold No Recorded Prime Contract Award.

What the War Department Put in Front of 1.7 Million Users on Aug. 31

The Department of War launched OpenAI’s ChatGPT Mil and Starshield AI’s Grok for Government on GenAI.mil on Aug. 31, 2026. Both are accredited for controlled unclassified information, or CUI, at Impact Level 5, the DoD tier one step below classified. They join Google’s Gemini for Government, live since the platform’s December 2025 launch. GenAI.mil has onboarded 1.7 million of the Department’s 3 million-plus personnel in nine months.

Your exposure runs two ways. The first is your Department-facing revenue: any generative AI seat, chat interface, retrieval layer, or document-processing tool at Impact Level 5. Your customer already reaches a substitute with a Common Access Card. How hard that squeezes you turns on a figure nobody has published: whether commands are charged back for seats. Free seats make this an option-year conversation, so ask your program office what a seat costs the command.

• What changed on Aug. 31 is narrower than it looks. A CUI-accredited substitute has sat behind that card since December 2025. The additions bought model choice, not the substitute’s arrival. If you were going to lose a seat-based line here, that pressure started nine months ago.
• The second is delivery. The Department named planning, policy, logistics, and administration as the workloads these tools support. For services firms billing hours in these workflows, the productivity baseline likely shifted after the platform became available in December 2025.
• Access is gated by equipment, not badge type. Contractor personnel on government-furnished equipment are authorized users. Your proposal team on corporate laptops is not.
• The platform is not authorized for personally identifiable information, protected health information, or classified material, so that work stays where it was.

The $600 Million in Ceilings That Produced Almost No Award Records

In July 2025 the Chief Digital and AI Office gave prototype other transaction agreements to four companies: OpenAI, Google, xAI, and Anthropic. Each carries a ceiling of up to $200,000,000 over two years. Those are maximums if every option is exercised, not money that has moved. Google is set aside below, since Google Public Sector carries unrelated federal work. xAI is scored as itself, the entity that signed. Its Grok now ships under the Starshield AI brand, and SpaceX does hold federal contracts.

• Anthropic, PBC: one recorded federal prime contract transaction, $18,960 obligated by the State Department on Feb. 11, 2026.
• OpenAI: no recorded prime contract award.
• xAI: no recorded prime contract award.

A Department pipeline should not be built around ceiling values that have produced no obligations. Two explanations fit. Prototype other transaction agreements are not FAR-based contracts, so they likely never reach the records a capture lead watches. And in four of the five awards below, the recipient of record is a reseller, not the model vendor. These are examples, not the full set.

Keep these as separate buying environments. Each agency runs its own appropriations, contracting officers, and vendor relationships. The Department of War is a fifth, and the only one where an accredited platform sits between buyer and vendor.

Who Is Already Positioned Between the Model and the Mission

The competitive ground is crowded, and none of it is the model layer. Prototype other transaction authority is the acquisition vehicle class most open to nontraditional entrants. The instrument is not what excludes you. Those four agreements are awarded and the tier is occupied. Treat it as taken, not closed.

• BigBear.ai bought Ask Sage for $250,000,000 in cash, private consideration rather than federal spending. That agreement predates the GenAI.mil launch and closed Dec. 31, 2025. Ask Sage served about 100,000 users across 16,000 government teams on roughly $25,000,000 of expected 2025 recurring revenue, by its own account. BigBear.ai now says it will retire that brand for Department of War customers.
• Government-built alternatives hold ground. The Air Force is sunsetting NIPRGPT. The Army kept CamoGPT and its Enterprise LLM Workspace. The Coast Guard, under Homeland Security, is enhancing Ask Hamilton.
• The channel tier carries the visible obligations. FCN, Inc., Carahsoft, and Software Information Resource Corp. are recipients of record on commercial AI license awards, and hyperscaler paths run through integrators.

No named competitive set exists yet for agent and mission-data work on the platform itself, which is both the risk and the reason to move now.

The fourth vendor is a warning rather than a competitor. A federal judge ruled on Aug. 27, 2026 that the Department’s designation of Anthropic as a supply chain risk was unlawful. Anthropic still is not on the platform, and a second suit is pending. A designation removes a vendor immediately; litigation does not restore the lost months. That asymmetry belongs in your risk register.

What GenAI.mil Consolidation Looks Like When It Reaches Your Contract

The Navy, the Air Force, and the Marine Corps have each designated GenAI.mil their enterprise generative AI platform. The Navy went furthest. A Jan. 28, 2026 memorandum made it the Department of the Navy’s enterprise IT service for CUI and Impact Level 5 generative AI. It directed commands to transition by Apr. 30, 2026, and to report every current and planned generative AI capability within 15 days.

That reporting requirement is the mechanism most likely to reveal where consolidation pressure reaches existing capabilities. It produces a list, and lists of duplicative capability get worked. Consolidation has not been uniform: the Air Force is retiring its own tool while the Army kept two.

How that pressure reaches a contract you hold:

1. A command inventories its generative AI capabilities and finds your tool on the list.
2. Your program manager is asked why a separately funded capability is needed alongside a platform at the same impact level.
3. That answer has to name what your tool does that the platform does not: proprietary data, a mission workflow, an accreditation boundary above Impact Level 5, an unreachable integration.
4. The conversation lands on an option exercise, not a termination notice.

No public record names a contractor generative AI contract terminated under these directives. The risk is a dated directive plus a named product line. It reaches you through budget justification rather than legal notice.

Where a Small or Mid-Size Contractor Can Still Transact

Only one of these three routes transacts today: the civilian channel. Agent development and accreditation work above Impact Level 5 sit at program formation, with no active procurement for contractor-addressable scope.

• Agent and mission-application development. The Department’s artificial intelligence strategy names Enterprise Agents as an enterprise-operations initiative. Personnel built 100,000 agents in April 2026, within roughly two weeks of that capability opening. That shows demand, and it shows government staff meeting that demand themselves without a contract. What sells is mission data, accountability for accuracy, and the agents nobody in uniform can build.
• Accreditation, IT modernization, and data work above Impact Level 5. The platform stops at CUI, so classified environments, protected health information, and personally identifiable information sit outside it.
• The civilian channel. Treasury, HHS, State, and smaller agencies still place reseller delivery orders for commercial AI licenses. The Department of War no longer does.

Take the step that fits:

• AI product firm or non-incumbent (DEFEND): price against the accreditation boundary above Impact Level 5, not the seat, and re-baseline any pipeline that assumed a seat-based sale.
• Mid-market systems integrator (SHAPE): Enterprise Agents requirements are not written yet. Take a position paper on mission-data accountability to the Chief Digital and AI Office while that holds.
• Small business DIB subcontractor (DEFEND): confirm whether on-site staff work on government-furnished equipment, which decides their platform access.
• Value-added reseller or distributor (WATCH): weight the civilian channel over the Department of War channel, on where obligations land.
• Agency-side technology owner: expect your component capability on the next inventory report, and decide now whether to defend it or fold it in.

The Two Expirations That Decide Whether This Holds

Two expirations carry this forward, controlled by different organizations. GSA’s governmentwide pricing term for Grok runs to March 2027. The Department’s own July 2025 prototype agreements reach two years in mid-2027. Neither governs the other.

A Department of War generative AI product line is currently a position to defend rather than expand aggressively. Put capture effort into the agent, data, and accreditation work the Department says it is building toward. Raise that priority if a component solicits agent work, or the capability extends to classified networks. Both would put competable dollars behind work you can bid. Lower it if a service reverses a directive, or the platform charges for seats.

FAQ

Which contracting shops should I monitor for work around GenAI.mil?

The Chief Digital and AI Office and the Office of the Under Secretary of War for Research and Engineering own the platform. Service CIO organizations own adoption and buy mission applications.

Which acquisition vehicle would carry agent or mission-application work here?

None is identified. The July 2025 prototype agreements cover the model tier, and neither Aug. 31 release names a delivery instrument. Nothing has reached solicitation, so this is a shaping window, not a bid.

Who is positioned in the civilian channel for these products?

FCN, Inc., Carahsoft Technology Corp., and Software Information Resource Corp. hold commercial AI license delivery orders at Treasury, Health and Human Services, State, and the National Gallery of Art. Smaller obligations sit at USAID, Commerce, and the Federal Communications Commission.

Why does Anthropic have almost no federal contract awards despite a $200 million ceiling?

Because a ceiling on a prototype other transaction agreement is permission, not an obligation. Anthropic is also not on GenAI.mil, following a supply chain risk designation a federal judge ruled unlawful on Aug. 27, 2026.

What would invalidate the argument that this compresses the contractor market?

A component solicitation putting competable dollars behind contractor-built capability at the same impact level, a service reversing its directive, disclosure that commands pay for seats, or Navy inventories producing no reductions. The Army keeping CamoGPT is already a partial counterexample.

What are the next observable events worth watching?

The March 2027 expiration of GSA’s Grok pricing term, the mid-2027 two-year mark on the prototype agreements, any extension to classified networks, and the remaining Anthropic litigation.