The Annual CR Risk: How Continuing Resolutions Compound the AI Technology Transition Problem

BLUF — Bottom Line Up Front

Fiscal year transitions under continuing resolution authority are a structural feature of how the federal government funds itself — not an exceptional event. For most defense hardware programs, a CR imposes delay and inefficiency but is survivable. For AI and software programs, the damage is qualitatively different: interrupted development sprints lose more than time, and the prohibition on new starts can block OTA follow-on work and expansions that were not funded in the prior year. Program offices that treat a CR as a temporary pause rather than a structural risk find themselves behind when full-year appropriations arrive. The practical question is how to reduce that exposure now, under the FY2027 CR that runs through December 11, 2026, and before next October 1.

How a Continuing Resolution Works

A continuing resolution is a joint resolution of Congress that funds government operations when appropriations bills have not been enacted. Obligation limits are enforced through the Anti-Deficiency Act (31 U.S.C. § 1341 bars obligations exceeding the amount available in an appropriation), and OMB apportions funds within those limits. CRs typically do two consequential things for defense programs.

First, they set an obligation rate. CRs typically calculate funding from a reference level, multiplied by the fraction of the fiscal year for which the funds are made available. The current CR, P.L. 119-103, signed September 2, 2026, covers October 1 through December 11, 2026, a 72-day period, at rates based on FY2026 enacted amounts. For a hypothetical program with $120 million enacted in FY2026, that rate supports roughly $10 million per month of obligations, or about $23.7 million over the 72-day CR, regardless of its FY2027 President's Budget request. If the program planned a large contract award early in the year, that award is either deferred or restructured.

Second, they prohibit new starts. For DoD, the CR bars new production of items not funded for production in the prior year, increases in production rates above those sustained with prior-year funds, and the initiation, resumption or continuation of any project or activity for which appropriations, funds or other authority were not available in FY2026 (Section 102(a) of the FY2027 CR, as summarized by CRS). The test is whether the activity was funded, not whether the contract is new: a new award continuing a funded effort is generally permitted, while a genuinely new effort waits for full-year appropriations or an anomaly. "New start" is defined in each CR's text and implemented through OMB apportionment and DoD financial management guidance, and the boundary is not always obvious, but the practical effect is clear: anything the program intended to launch as FY2027 business is held until the CR ends or an anomaly applies.

Congress has relied on CRs for at least part of every fiscal year since FY1998; FY1997 was the last fiscal year with no CR enacted. Full-year CRs were enacted for FY2007, FY2011, FY2013 and FY2025, and FY2025 was the first year DoD itself operated under a full-year CR (P.L. 119-4). FY2026 began with a 42-day lapse in appropriations, October 1 through November 11, 2025, before a CR was enacted on November 12, 2025. Defense program managers who assume October 1 begins a fully funded year are planning against the historical exception, not the rule.

If You Are Already Under the FY2027 CR

The current CR expires December 11, 2026. If Congress has not enacted another CR or full-year appropriations by then, a lapse in appropriations beginning December 12 is possible, as FY2026 showed. Program offices should confirm now which actions must be obligated before expiration, and which depend on a follow-on CR. The FY2027 CR also includes a Section 127 anomaly allowing up to $2.85 billion of CR funds in Procurement, Defense-Wide for "National Security Systems". Whether a software-heavy national security system qualifies for that anomaly is a question for the comptroller.

Where AI and Software Programs Feel It Differently

Traditional defense hardware programs — a vessel, a ground vehicle, an aircraft platform — have development schedules measured in years and production rates that are relatively insensitive to a 30- or 60-day funding interruption. The production line does not stop; the workforce does not disband; the contractor absorbs the delay and adjusts the schedule. The cost is real but bounded.

AI and software programs are structurally different in three ways that interact badly with CR mechanics.

Development cadence is not bufferable the same way. Software acquisition under the DoD Software Acquisition Pathway (DoDI 5000.87, issued October 2, 2020 under the authority of Section 800 of the FY2020 NDAA, Pub. L. No. 116-92) is built around modern iterative development (for example, agile and DevSecOps), with new capabilities delivered to operations at least annually and more frequent releases encouraged. A two-month funding interruption does not simply delay delivery by two months; it disrupts active sprints, forces the contractor team into a holding pattern, fragments institutional context built up across months of work, and frequently triggers personnel turnover as commercial-market engineers find other engagements. When the CR lifts, the program does not resume from where it stopped — it reorients.

New starts include OTA phase transitions. Other Transaction Authority agreements under 10 U.S.C. § 4022 are structured as prototyping agreements and follow-on production arrangements, not as indefinite-delivery vehicles that renew automatically. A prototype OTA that completes its first phase and is ready to transition to a follow-on production contract or a second prototype effort generally needs a new award. Under a CR, that award can be blocked if the follow-on work or expansion was not funded in the prior year. The OTA handshake — fast and flexible in normal appropriations — becomes a waiting game during a CR, and the contractor's team and test environment may not hold.

Cloud and data contracts require early-year obligation. AI programs increasingly depend on commercial cloud hosting, training compute, and data pipeline infrastructure provisioned through contracts with variable obligation requirements. Enterprise cloud capacity — for example, task orders on DISA's Joint Warfighting Cloud Capability vehicle (awarded to AWS, Google, Microsoft and Oracle, with funds obligated order by order) — needs obligation authority early in the year to keep services and licensing continuous and to add capacity ahead of fielding milestones. Recurring orders for funded, ongoing use are generally continuations, but a material expansion or a new environment for a new effort can raise a new-start question.

Three Failure Modes Under a CR

Experience across programs that have navigated CRs with AI or software-intensive work reveals three recurring failure modes.

The "pause-then-sprint" miscalculation. A program office, anticipating that the CR will lift by December, shifts all major planned activities to the back half of the fiscal year. When the CR runs longer than expected — or when full-year appropriations arrive but with conditions that require re-planning — the program compresses eight months of work into six. Sprint-based teams under compressed timelines make different quality tradeoffs than teams with realistic schedules. This is an analytical risk, not a documented GAO finding. GAO's 2021 review of DoD's CR management (GAO-21-541) found that the programs it reviewed avoided delays or cost increases, with the services starting service contracts after the first quarter and postponing nonessential purchases to later in the fiscal year, the same back-loading this failure mode describes. The risk is that back-loading works only while CRs end on schedule; when a CR runs long, the deferred work has less room to land.

OTA team dissolution. For programs using OTAs, contractor teams assembled for a prototype phase are often composed partly of specialists hired specifically for that engagement. If a CR-induced pause stretches toward a quarter or more, contractors in the defense technology industrial base — particularly smaller firms and non-traditional vendors — face a choice between holding unproductive staff and releasing them. The talent goes elsewhere. The program may formally restart on time after the CR ends, but with a meaningfully different team than the one that built the Phase 1 capability.

Missed testing windows. AI systems integrated into platform programs — targeting systems, ISR processing layers, logistics optimization tools — typically have test event windows tied to platform schedules, range availability, and operational unit availability. A CR delay of two months does not necessarily move a test window by two months; it may move it by many months, to the next available range slot that aligns with all the required constraints. Programs that fall out of their testing windows often restructure delivery schedules in ways that cascade into subsequent milestones.

What Program Offices Can Do Before October 1

The CR is not fully preventable, but much of its impact on AI and software programs can be mitigated with pre-year-end actions.

Obligate what can properly be obligated. Before September 30, award modifications, task orders and option exercises for requirements that are bona fide needs of the current fiscal year (31 U.S.C. § 1502 makes expiring funds available only for expenses properly incurred during the period of availability), including severable services that start this year and run no longer than 12 months (10 U.S.C. § 3133). Coordinate with the comptroller and counsel on what qualifies. A program that properly anticipates its first-quarter needs in Q4 preserves its development cadence through the CR period at minimal overhead.

Negotiate bridge provisions into OTA structures. When structuring prototype OTAs, include explicit bridge funding and continuation authority for defined transition scenarios. Structure the agreement so the follow-on effort is clearly the continuation of a funded project, and confirm with the comptroller before October 1 whether the transition counts as a new start. An option clause speeds award once funds are available; it does not by itself satisfy the new-start test.

Document new-start thresholds explicitly. Work with the program's legal counsel and comptroller to establish a written agreement with the service on which upcoming actions qualify as new starts and which are continuations. This agreement has no legal effect on the CR itself, but it provides the administrative record needed to move quickly when the CR lifts — and it surfaces ambiguities early enough to address them.

Maintain contractor technical teams. For programs with non-traditional vendor teams, build CR continuation provisions into the contract as firm base work rather than unexercised options. A team that has been retained and oriented can execute from the day the CR ends. A team that has disbanded can take months to reconstitute, regardless of the technical requirements of the restart.

Develop a CR playbook. Programs that enter each October with a documented, rehearsed CR management plan — pre-staged obligations, identified bridge provisions, cleared ambiguities — navigate the transition significantly more smoothly than programs that treat each CR as novel. The playbook should be owned by the program executive and reviewed with the contractor at the end of each fiscal year.

The Structural Problem Behind the Annual Pattern

The root issue is not complexity in any specific rule but a mismatch between the federal budget process — designed for hardware programs with production schedules measured in years — and software acquisition architectures that require consistent funding to maintain velocity. Congressional direction through successive NDAA cycles established and expanded the Software Acquisition Pathway — now codified at 10 U.S.C. § 3603 — precisely because traditional acquisition frameworks were poorly adapted to modern software delivery. The CR problem is one of several places where that mismatch shows up most clearly.

Until congressional appropriations processes align more closely with continuous development cadences — which would require authorities and mechanisms that do not currently exist at scale — defense AI and software programs will continue to absorb disproportionate CR damage relative to traditional programs. The mitigation approaches above reduce that damage; they do not eliminate it.

Spartan X's program execution and acquisition support work — scheduling, risk management, and FAR/DFARS contract structuring — is where this planning happens. Programs that treat budget uncertainty as a design input, and structure their contracts, OTAs and teams accordingly, give themselves options that programs planning for continuous appropriations do not have.