Department of Defense scales up AI lab contracting

Last updated
The Department of Defense (DOD) quietly begins contracting OpenBrain directly for cyber, data analysis, and R&D, but integration is slow due to the bureaucracy and DOD procurement process.

Late 2026: AI Takes Some Jobs · Archived PDF, p. 8

Direct frontier-lab defense contracting began before the late-2026 scenario window. The contracting milestone is observed early; award ceilings do not establish expenditure or completed operational integration.

At a glance

  • Assessment: Ahead
  • Confidence in assessment: 95%
  • Outcome: observed
  • Timing: early
  • Predicted timing: Late 2026
  • Adjusted timing: Mid-2025 (June to July 2025 awards; about 16 to 18 months ahead of the late-2026 window)
  • Primary source: Late 2026: AI Takes Some Jobs, archived PDF p. 8

What AI 2027 Predicted

The Department of Defense (DOD) quietly begins contracting OpenBrain directly for cyber, data analysis, and R&D, but integration is slow due to the bureaucracy and DOD procurement process.

Source: Late 2026: AI Takes Some Jobs: archived PDF, p. 8.

Footnote 23 specifies an Other Transaction Authority (OTA) contract with a high-priority DX rating. The authors say they chose this specific mechanism for concreteness and express uncertainty about its likelihood; the contract is publicly announced but not emphasized in OpenBrain communications.

How We Track This

We monitor:

  • DOD contracts with frontier AI labs (OpenAI, Anthropic, Google DeepMind)
  • CDAO (Chief Digital and AI Office) initiatives and procurement
  • Defense tech partnership announcements
  • Congressional hearings and appropriations related to AI
  • Integration challenges and bureaucratic friction reports

Current Evidence

CDAO announced awards to OpenAI, Anthropic, Google and xAI on July 14, 2025, each with a $200 million ceiling. This establishes contracting before the late-2026 window, not $800 million already spent. CDAO announcement · DefenseScoop, July 14, 2025

This prediction has materialized ahead of schedule, though the process has been more politically contentious than the scenario anticipated. Ahead rests on contract timing, not on integration speed.

Anthropic’s documented $200 million-ceiling agreement was announced July 14, 2025. Anthropic explicitly describes it as a two-year prototype other transaction agreement, not $200 million of realized spending. This verifies the OTA-prototype mechanism for Anthropic’s award; the cited company announcement does not establish that all four July awards used that instrument or that any carried the scenario’s DX rating. The later 2026 political and contractual disputes are separate developments. Anthropic

  • Anthropic was effectively blacklisted after refusing to remove safety guardrails at the Pentagon’s request. Trump directed all federal agencies to “immediately cease” use of Anthropic technology. CNBC, March 26, 2026 reports the directive was issued in late February.
  • OpenAI stepped in to sign its own agreement with the Department of Defense (now rebranded “Department of War”), publicly announcing the deal in late February 2026.
  • Significant backlash followed: 30+ OpenAI and Google DeepMind employees filed statements supporting Anthropic’s lawsuit against the government.

The scenario’s prediction of DOD-AI lab contracting scaling up is directionally consistent with what has occurred, though the mechanism differs significantly. The predicted slow due to bureaucracy friction has instead taken the form of political controversy and inter-lab competition: a qualitatively different dynamic than administrative delay.

In May 2026, TechCrunch reported that the Defense Department signed additional agreements with Nvidia, Microsoft, AWS, and Reflection AI after earlier deals with Google, SpaceX, and OpenAI. The agreements allow deployment of AI technology and models on classified networks for “lawful operational use,” including IL6 and IL7 environments, and the DOD said the architecture is intended to avoid AI vendor lock-in. This strengthens the evidence that frontier-lab defense integration is broadening from prototype contracting into higher-sensitivity deployment environments.

AP’s May 2026 coverage of the Anthropic-Pentagon appellate dispute reinforces the caveat that DOD-frontier-lab integration is scaling amid legal and political friction. This is not new evidence for contract scale by itself; it is useful counterevidence against a simple “quiet integration” narrative.

On June 5, 2026, the White House issued NSPM-11 on national security AI. It directs the national security enterprise to accelerate adoption, maintain proactive industry partnerships, make advanced frontier models broadly available to national security professionals, rapidly onboard advanced AI models from multiple vendors, and build next-generation high-security compute facilities. This further supports the already-confirmed defense-AI integration trend, while still not proving the depth of operational use inside specific military workflows.

Sources:

Counterevidence & Limitations

  • The scenario described this as quiet: the reality has been anything but, with massive public controversy
  • The scenario assumed a single dominant lab (OpenBrain) whereas reality involves multiple competing labs with different stances on military work
  • Actual integration into DOD workflows remains unclear: contracts signed ≠ capabilities deployed
  • The political dimension (Anthropic blacklisting, employee revolt) was not anticipated in the scenario
  • It remains unclear how much of the contracted capability is for cyber and R&D vs. administrative/analytical use

What Would Change Our Assessment

Maintain Ahead: contracting with multiple frontier labs continues ahead of late 2026. Move to Confirmed with early timing: if the site scores the contracting element alone as observed. Weaken: contracts cancelled or not renewed without replacement.

Update History

DateUpdate
2026-11-07Assessment reviewed. Status: Ahead; confidence: 95%. Anthropic’s July 14, 2025 announcement documents a two-year DOD prototype agreement with a $200 million ceiling before the late-2026 window. The contracting milestone occurred early; the ceiling is not spending, and the announcement does not establish completed integration or a DX rating. Correction: Direct awards began in June and July 2025, before the late-2026 window. Anthropic’s July award was a two-year prototype OTA with a $200M ceiling; the evidence does not establish a DX rating. Corrected the Anthropic award date, the subject of its 2026 impasse and attribution of the federal stop-use directive. March 26 reporting establishes a preliminary injunction, not the separate April temporary restraining order previously described.
2026-09-07Changed confirmed to ahead to reflect the already-observed early contracting milestone. July 2025 awards preceded the late-2026 window; operational integration remains qualified. Confidence remains 0.95.
2026-09-06Current evidence and assessment explanation reviewed. Clarified measurement scope, source interpretation and limitations.
2026-06-08The White House issued NSPM-11 on national security AI, directing accelerated adoption and proactive industry partnerships to make advanced frontier models available to national security professionals. This further confirms broadening defense-AI integration, though operational depth remains unclear.
2026-05-25AP reported that appellate judges appeared divided in Anthropic’s legal dispute with the Pentagon over its supply-chain-risk designation. This reinforces that DOD-frontier-lab integration is scaling amid legal and political friction rather than as a quiet procurement process.
2026-05-04Pentagon agreements expanded to Nvidia, Microsoft, AWS, and Reflection AI after earlier deals with Google, SpaceX, and OpenAI, allowing AI deployment on classified IL6/IL7 networks for lawful operational use. This further confirms defense-AI integration is scaling ahead of the scenario’s late-2026 timing, while the exact operational depth remains unclear. Confidence adjusted 0.90 → 0.95.
2026-04-20Reuters reported on April 16 that Google was negotiating a Pentagon agreement that would allow Gemini models to be deployed in classified settings, according to The Information. That is fresh evidence that frontier-lab defense contracting is spreading beyond OpenAI and into higher-sensitivity environments, reinforcing that DOD adoption is broadening across multiple major labs rather than remaining a one-company dynamic.
2026-04-06Federal judge issued temporary restraining order blocking DOD from enforcing Anthropic exclusion from Pentagon procurement; government immediately appealed (WebProNews). Project Maven designated as official “program of record” with US Army managing all future contracts (Dataconomy). Palantir distanced itself from AI-driven targeting deaths. Harvard Berkman Klein Center held event on “Agentic AI and Cybersecurity: Threats, Governance, and Strategy” (Apr 3). The legal and ethical battles around military AI contracting continue to intensify — consistent with the scenario’s “slow due to bureaucracy” friction, though manifesting as political/legal disputes rather than administrative delay. No status change. Correction (November 7, 2026): the WebProNews link above is an archived historical citation, not verified current evidence. CNBC, March 26, 2026 directly reports a preliminary injunction; it does not verify a separate April 2026 temporary restraining order. Use that report for the verified ruling, retaining the original wording above as history.
2026-03-30Pentagon blacklisted Anthropic as a “supply-chain risk to national security” after Anthropic refused to remove clauses prohibiting mass domestic surveillance and fully autonomous weapons from its DOD contract. Defense Secretary Hegseth cancelled Anthropic’s $200M Pentagon contract and directed all federal agencies to stop using Anthropic technology. OpenAI simultaneously struck a deal with the DOD under an “all lawful purposes” framework with architectural controls (cloud-only, cleared engineers embedded forward). Anthropic CEO Dario Amodei called the action “retaliatory and punitive” and promised legal challenges. Legal scholars question whether the “supply-chain risk” designation is legally sound when applied to a US company over a commercial dispute (Malwarebytes). This development is consistent with the prediction’s framing of DOD scaling up contracting with the “winner” of the AI race — though the mechanism (competitive displacement rather than parallel contracts) wasn’t anticipated. No status change. 2026-11-07 · Correction: The federal-wide stop-use directive was issued by President Donald Trump, not Defense Secretary Hegseth; February 28 reporting distinguishes the presidential directive from the Pentagon action.
2026-03-23Pentagon designates Palantir’s Maven Smart System as official “program of record” per Deputy SecDef Feinberg memo (Mar 9), locking in long-term AI integration for warfighter decision-making across all domains (Reuters). Separately, Reuters reports Pentagon official warned AI contract restrictions could “paralyze US military missions” (Reuters). Lawfare analysis details procurement complexity of military AI contracts. No status change — further confirms DOD-AI integration scaling ahead of the scenario’s late-2026 timeline.
2026-03-16Anthropic sues Pentagon (Mar 9) over “supply chain risk” designation, claiming First Amendment violation. 30+ OpenAI/Google employees file amicus brief supporting Anthropic. OpenAI simultaneously admits Pentagon tested its models via Microsoft even during its prior military-use ban. WIRED report deepens picture of competitive dynamics. No status change — confirms ongoing escalation.
2026-03Significant political drama around OpenAI-DOD partnership not anticipated by the scenario. Contracting scale and speed exceed predictions, but bureaucratic integration challenges remain as predicted.
2026-02Anthropic blacklisted from new defense contracts over safety testing timeline dispute. OpenAI signed DoD agreement same month, detailing “layered protections.” The predicted military-AI integration is happening, but the lab-Pentagon relationship is more fractured than the essay depicted. 2026-11-07 · Correction: Anthropic’s February 27, 2026 statement describes an impasse over exceptions for mass domestic surveillance and fully autonomous weapons, not a safety-testing timetable.
2026-01Pentagon released “Artificial Intelligence Strategy for the Department of War” — declaring 2026 the “year of Military AI Dominance,” directing “wartime speed” AI adoption, and requesting a record $14.2B for AI and autonomous research in FY2026.
2025-12Department of Defense signs $200M+ contracts with multiple AI labs, arriving ahead of the predicted timeline.
2025-11Trump signs the Genesis Mission executive order (November 24), directing federal agencies to identify 20+ science and technology challenges addressable via AI. Embeds AI in the federal R&D mission broadly. Builds on existing $200M Pentagon contracts. This prediction has arrived approximately 18 months ahead of the AI 2027 timeline.
2025-07Pentagon formally announces four $200M contracts on July 14: Anthropic, OpenAI, xAI, and Google each receive identical two-year “agentic AI” agreements. The explicit use of “agentic AI” in the announcement framing is notable.
2025-06US DoD CDAO awards Anthropic a two-year $200M ceiling prototype contract for “frontier AI capabilities advancing US national security.” OpenAI receives a comparable contract. This is approximately 18 months ahead of AI 2027’s predicted timeline for government pulling AI labs into defense-contractor relationships. 2026-11-07 · Correction: OpenAI announced its CDAO pilot on June 16, 2025; CDAO announced Anthropic, Google and xAI alongside OpenAI on July 14, 2025. The original row misdated Anthropic’s award.