The Software Appropriation Problem the FY2027 NDAA Could Finally Fix
Back to Signal
AIDefenseGovernmentInnovation

The Software Appropriation Problem the FY2027 NDAA Could Finally Fix

November 22, 2026Spartan X Team

The Structural Mismatch

The Department of Defense appropriates funding through discrete categories, each with rules about what it can and cannot fund. The main categories relevant to software are Research, Development, Test and Evaluation (RDT&E), which funds development activity including research, prototyping, and test and evaluation before a system reaches production; Procurement, which funds one-time acquisitions of hardware, infrastructure, or software that become capital assets; and Operations and Maintenance (O&M), which funds the operation and sustainment of fielded capabilities.

For traditional defense programs, this structure aligns reasonably well with the acquisition lifecycle. A system is developed under RDT&E, acquired under Procurement, and sustained under O&M. Each appropriation category corresponds to a distinct phase, and transitions between them coincide with milestone decisions.

Software programs do not follow this lifecycle. Under the Software Acquisition Pathway (SWP), established by FY2020 NDAA Section 800 (now codified at 10 U.S.C. §3603 by FY2025 NDAA §805) and made the preferred pathway for all software development in business and weapon system programs by a March 6, 2025 Secretary of Defense directive, a program moves from planning to execution and then delivers continuously. DoDI 5000.87 §3.3 requires, for applications programs, the minimum viable capability release within one year of first obligation and further releases at least annually — timelines that mirror the considerations Congress set in 10 U.S.C. §3603(d) — with more frequent delivery where risk and test strategies support it (embedded software follows the host system's delivery schedule). The whole model assumes that "development" and "sustainment" are not sequentially distinct phases but parallel, continuous activities.

That assumption collides with the appropriation categories. When a software program in execution is delivering monthly releases, which appropriation type applies to its work? If RDT&E, the program is in perpetual development — a characterization difficult to sustain as the system matures. If O&M, the program cannot fund development activity. If Procurement, each iteration must be justified as an investment acquisition rather than continuing work. None of these mappings is accurate for a program running continuous delivery.

The practical result is a set of workarounds that have become normalized. Programs label ongoing improvement activity as "development" to stay in RDT&E beyond the period it accurately describes. Others create artificial version milestones — declaring a new major version a fresh development program — to obligate Procurement funds for continued evolution. Others split their budget exhibits: one line funded under RDT&E for new features, another under O&M for maintenance, with a boundary that does not correspond to how the software is actually built or tested. One partial mitigation already appears in DoD budgets: the Software and Digital Technology pilot budget activity, which funds a limited set of software programs through a dedicated budget line. It does not extend to the broader program population or to AI programs outside it.

The Commission on Planning, Programming, Budgeting, and Execution Reform, created by FY2022 NDAA Section 1004 (Public Law 117-81) and completing its work in March 2024, examined this structural problem as part of its comprehensive review of the PPBE system. H.R. 8800 Section 1521 directs that the revised guidance maintain consistency, to the maximum extent practicable, with Recommendation 11A of the Commission's final report.

These workarounds carry compliance risk. Appropriations law requires that funds be used for the purposes for which they were appropriated. A program that chronically characterizes sustainment activity as development, or development activity as sustainment, to stay inside a convenient appropriation category is managing legal risk as much as budget risk. The workaround is a symptom of the mismatch, not a durable solution.

What Section 1521 Would Change

House FY2027 NDAA Section 1521 addresses the mismatch at the regulatory level. It would insert a new 10 U.S.C. §2221 — a section number vacant in Chapter 131 since a 1998 repeal, to be titled "Availability of appropriations accounts for full lifecycle of software capabilities: regulations" — directing the Secretary of Defense to ensure the financial management regulations allow any primary appropriation funds to be applied flexibly across the full lifecycle of software capabilities, without regard to appropriation category. The revised regulations must be issued within one year of enactment.

The specific statutory mechanism is a revision requirement, not a direct appropriation. Congress is not creating a new account or changing what it appropriates; programs would still receive RDT&E, O&M, or Procurement funds in their budget exhibits. What would change is the internal DoD regulatory constraint that restricts how those funds are applied within the software lifecycle. After the FMR revision, a program funded under O&M could use those funds for feature development activity that the current regulation treats as RDT&E-exclusive — and a program funded under RDT&E could apply those funds to sustainment activities without constructing a fiction about their nature.

The authority is specifically scoped to software capabilities. It does not change appropriation rules for hardware, weapons systems, or non-software programs. It also does not eliminate budget transparency or audit requirements — programs will still need to account for how funds are spent, just without the artificial category constraint on how they are deployed.

What Section 1521 Does Not Do

The provision would not change the appropriations process, the category structure Congress uses, or the fiscal law framework that governs appropriate use of appropriated funds. GAO's role in determining whether DoD spent funds for the purposes appropriated remains unchanged. If Congress appropriates RDT&E funds for a specific program line, the program still needs to use those funds in ways consistent with congressional intent; the FMR revision creates regulatory flexibility within that constraint, it does not override statutory direction.

The provision also does not take effect on enactment. It requires the Secretary of Defense to revise the Financial Management Regulation within one year. Service- and program-level implementation guidance is likely to follow the regulation itself, not accompany it. Programs should not expect to file amended budget exhibits the week an NDAA is signed.

There is also no guarantee of continuity across budget years. Even with a revised FMR, programs still present budget exhibits by appropriation category and will need to explain their approach to program executive offices and comptroller staffs that have not yet adapted to the new framework. The revision creates the authority; individual programs create the execution discipline that uses it responsibly.

The Senate Bill and the Conference Decision

The Senate Armed Services Committee voted 18-9 on June 10, 2026 to advance its version of the FY2027 NDAA. The committee's executive summary describes its cybersecurity and AI provisions as including codifying autonomous weapons review in statute, directing a DoD-wide agentic AI ecosystem, establishing AI agent security standards, and assessing AI effects on warfighter skill retention. S. 4784 contains no provision comparable to House Section 1521. Its closest software item, Section 802, withholds some funds pending a report on Software Acquisition Pathway implementation.

Senate floor consideration of the bill stalled in July 2026, leaving the two chambers to negotiate in the lame-duck session. Conference negotiations routinely add, modify, and drop House-only or Senate-only provisions. Section 1521 has no Senate conferee champion established from the available record and no matching Senate provision to merge with — which puts it in the pool of provisions that need active negotiation rather than automatic reconciliation. Its survival depends on whether House conferees make it a priority and whether Senate conferees accept it.

This is real conference risk. The provision is specific and consequential, but House-only provisions with no Senate analog face the full weight of negotiation tradeoffs in a lame-duck session under time pressure. Program offices should track the conference outcome before restructuring their budget strategies around the new authority.

Why This Matters More for AI Programs

AI programs on the Software Acquisition Pathway face the appropriation mismatch in an acute form. The standard software lifecycle already creates category confusion; a machine-learning program's lifecycle layers additional complexity on top of it.

Consider, as a hypothetical illustration, a DoD logistics AI program on the applications path with monthly release cycles. Training the initial model is unambiguously development — RDT&E. Running inference on production data is unambiguously operations — O&M. But retraining the model on updated operational data, to prevent drift as conditions change, sits between the two. It uses the same computational infrastructure as initial training. It changes the model's behavior. It is not a new capability — it is maintaining the existing one. Under current FMR guidance, the program must characterize this activity as either development or sustainment, and neither is fully accurate.

The same ambiguity applies to monitoring and drift detection (sustainment that informs development decisions), A/B testing of model variants in production (development activity running in the operational environment), and annotation and data-quality work that feeds the training pipeline (a development activity that cannot stop during sustainment). AI programs that follow sound engineering practice run all of these activities continuously. The appropriation categories were written for systems where these activities had natural phases; they do not map cleanly onto a continuous ML lifecycle.

The SWP AI governance considerations examined in depth earlier this fall — model versioning, evaluation currency, drift detection, and the oversight artifacts that hold AI governance — remain necessary regardless of this provision. But the appropriation flexibility would remove one of the structural constraints that forces AI programs to paper over the continuous nature of their work. A program with the authority to apply O&M funds to model retraining does not need to maintain an artificial distinction between its "sustaining" and "developing" functions. It can design a single continuous process and fund it accordingly.

What to Do Before the Conference Concludes

There is no reason to wait for enactment to understand the authority and its implications.

Map where current budget exhibits create artificial phase breaks. Identify the activities that the program labels as one appropriation type for budget purposes but that functionally span multiple types. These are the locations where the new authority would apply and where current workarounds carry compliance risk. Document them before the conference outcome is known.

Model what a revised budget exhibit would look like. If the FMR permitted flexible application of any primary appropriation across the full software lifecycle, how would the program's FY2029 exhibit look different from its current structure? This is a planning exercise, not a legal commitment. Doing it before enactment means the program has a template ready rather than starting from scratch after the regulation changes.

Identify which activities would need re-categorized FMR guidance. The FMR revision will take time and will have specific scope. Programs that have catalogued their edge cases will be better positioned to submit feedback on draft guidance, raise edge cases with their component comptroller staffs, and avoid surprises when the guidance is finalized. Outside the limited Software and Digital Technology pilot budget activity, the FMR has no general framework for continuous-delivery software funding; the revision will need to establish language for activities that have no clean current category.

Review existing workarounds for continued validity. A revised FMR will not simply validate every current practice. Some workarounds that addressed the category mismatch will become unnecessary; others may conflict with the new framework's specific requirements. Programs that have been stretching one appropriation category to cover activities it was not designed for should review whether their current approach will be absorbed, superseded, or invalidated by the new framework.

Do not front-run the conference. A House-passed bill is not enacted law. Briefing acquisition leadership on what the authority would mean is reasonable. Restructuring budget exhibits or reprogramming on the assumption of enactment before the conference agreement is announced is not.

The provision is specific and addresses a problem that has been documented since the Software Acquisition Pathway launched. Whether it survives conference is not predictable. What is predictable is that programs that understand it now will be better positioned than those that encounter it for the first time after enactment.

Spartan X works with federal program offices on acquisition strategy and the financial management frameworks that govern software and AI programs. Getting ahead of a regulatory change before the guidance is written — mapping current structures, modeling alternatives, and identifying edge cases — is exactly the kind of problem our teams address with program leadership.

Sources and further reading

Share this article
LinkedIn

RELATED

More from Signal

The 618th AOC's $26M Flyways Purchase: What Sole-Source Commercial AI at the Global Mobility Hub Looks Like
November 21, 2026

The 618th AOC's $26M Flyways Purchase: What Sole-Source Commercial AI at the Global Mobility Hub Looks Like

The Air Force awarded a $26 million sole-source purchase order for Air Space Intelligence's Flyways AI platform to the 618th Air Operations Center — the command that plans and executes every U.S. strategic airlift and air refueling mission globally. The acquisition pattern behind it, pilot followed by no-compete scale, is repeating across operational defense AI, and the data rights, security integration, and exit strategy questions it raises rarely get answered before the first check clears.

Read More
The Army's TITAN Production Award Is Smaller Than You Think — and That's the Point
November 19, 2026

The Army's TITAN Production Award Is Smaller Than You Think — and That's the Point

The Army announced production contracts totaling $192M for eight TITAN AI-enabled ISR ground stations on September 1, 2026 — a small initial tranche while a larger follow-on order is anticipated in FY27, likely before most of those eight systems are even delivered. What the initial production actually buys is not a waiting period; it is the operational infrastructure the Army needs to absorb a production fleet at all.

Read More
The DoW Super Intelligence Office: What the CDAO Rebrand Signals for Defense AI Buyers
November 18, 2026

The DoW Super Intelligence Office: What the CDAO Rebrand Signals for Defense AI Buyers

On November 13, the Department of War redesignated its Chief Digital and Artificial Intelligence Office as the Super Intelligence Office. The name follows a September 29 executive order (EO 14434), but the architecture underneath—Maven, War Data Platform, GenAI.mil, Enterprise Agents—is unchanged. Here's what the transition means for programs already under contract and for contractors entering the pipeline.

Read More

BUILD WITH US

Ready to Solve Hard Problems?

Spartan X builds AI systems, autonomous platforms, and cybersecurity solutions for defense and national security.