By Indunil Ranaviraja, Head of Mission Enablement & Innovation, REI Systems
At the 2026 AFCEA Bethesda Health IT Summit, federal health leaders signaled a clear shift: AI is no longer just an adoption priority. It is an accountability and change management challenge — and in federal health, both trace directly to the people these systems serve and the public servants who operate them every day.
Agencies are already using AI to accelerate analysis, reduce repetitive work, strengthen fraud detection, support decision-making, and manage growing operational complexity. The question now is whether AI can be governed in a way that matches the consequences it carries and whether the people most affected by those consequences are kept at the center of the design. Getting that right is not only a governance problem. It is a change management problem. Technology governance without aligning with workforce adoption maturity, institutional knowledge, and human-centered design does not deliver mission outcomes. It delivers technical compliance.
That distinction matters enormously in federal health. AI is being considered for workflows tied to coverage determinations, eligibility checks, fraud prevention, regulatory review, acquisition, program integrity, workforce productivity, and public-facing service delivery. At the end of every one of those workflows is a beneficiary, a patient, a provider, a caregiver, or a public servant trying to do their job well. When AI gets those workflows right, it delivers faster, more accurate service to real people. When it gets them wrong, or when it is governed poorly, adopted without workforce readiness, or deployed without meaningful human oversight. real people bear the cost.
The directional answer is not to slow federal health AI adoption, but to connect every AI use case to a clearly defined health mission outcome, a named accountable owner, and a human-centered operating model before it scales. Agencies should begin with high-value, bounded workflows where better data, better decisions, and better oversight can directly improve service to patients, beneficiaries, providers, and program staff.
The Threshold That Matters Most
One of the most important governance questions is not simply whether an agency is using AI. It is what role AI plays in the workflow and who is downstream of that role.
There is a meaningful difference between AI that helps a person draft, summarize, search, or organize information and AI that materially shapes a decision, triggers an action, prioritizes a case, flags a transaction, or influences an outcome. That threshold is where oversight requirements increase, change management complexity grows, and the human stakes rise.
OMB Memorandum M-25-21, “Accelerating Federal Use of AI through Innovation, Governance, and Public Trust,” draws this exact line into federal policy. It designates AI as “high-impact” when its output serves as a principal basis for decisions or actions with legal, material, binding, or significant effect, and it names healthcare, including diagnosis, treatment, and public insurance, as a presumed high-impact category. This is not an abstract governance concept. It is the operative federal standard already governing the coverage, eligibility, and fraud-related workflows this piece describes.
At that threshold, agencies need stronger controls: explainability, auditability, pre-deployment testing, ongoing monitoring, human review and appeal rights, and the incorporation of end-user feedback are the minimum risk management practices that M-25-21 requires for high-impact AI use cases, in addition to performance monitoring, escalation paths, and lifecycle management. A fraud flag, clinical review recommendation, regulatory prioritization cue, or eligibility-related workflow does not carry the same risk profile as an internal productivity tool. The difference is not technical. It is human. And the change management requirements at that threshold, such as workforce training, process redesign, adoption support, and trust-building with end users, are as important as the governance architecture itself.
The results of getting this right are already visible. According to Federal News Network, Bethany Messick, acting deputy director of the CMS Center for Program Integrity, said the agency’s Fraud Defense Operations Center had assessed approximately 300 Medicare providers and recovered about $2 billion in fraudulent payments. Separately, Nextgov/FCW reported that CMS Deputy Administrator and Chief Operating Officer Kim Brandt said the use of AI in fraud analyses enabled CMS to analyze claims before payment and helped the team save $2 billion since March 2025. Both accounts reinforce the importance of human oversight: Messick described investigators, policymakers, data scientists, and legal counsel working together, while Brandt said humans continue to double-check AI-supported analysis for accuracy. The outcomes did not come from AI alone. They came from technology operating within a model where human judgment remained central.
Agencies need governance models that reflect the consequences of the use case, and that ask explicitly: who is affected when the AI gets it wrong, and are the people operating these systems truly equipped to catch it?
The solution is risk-tiered AI governance: classify health AI use cases by consequence, then apply proportionate controls for explainability, human review, appealability, monitoring, and escalation. For high-impact health workflows, agencies should design the governance model around the patient, beneficiary, provider, or program decision at stake and not just around the model or tool being deployed.
AI Economics Are Now Part of the Governance Conversation
The summit conversations surfaced a challenge that deserves more direct attention: as agencies scale AI, the economics of operating it become a governance question, not just a budget one.
Compute costs, token consumption, licensing models, data infrastructure, and cloud usage can escalate quickly without visibility into how AI is being used. Agencies applying AI at enterprise scale need to understand not only whether AI works, but whether it can be operated and sustained responsibly, and whether the resources invested are being directed toward programs and populations that need them most.
That makes FinOps and Technology Business Management increasingly relevant. AI governance cannot focus only on model performance, ethics, and oversight. It also must address cost transparency, usage accountability, value realization, and investment discipline. OMB Memorandum M-25-22 already makes this a procurement requirement, not just a best practice it directs agencies to build ongoing performance and cost-effectiveness monitoring into AI contract terms across the acquisition lifecycle.
For federal health leaders managing large portfolios where program funds are public resources tied to public missions, this is critical. AI can create meaningful efficiency gains, but current federal financial management structures were not built for the cost dynamics of AI at scale. Visibility into token consumption, compute costs, and usage patterns across programs requires investment in new tools and reporting disciplines that most agencies are still exploring. The agencies getting ahead of this are not waiting for perfect visibility. They are starting with the use cases that carry the most consequence and building cost accountability there first. Responsible AI is not only about trust. It is also about sustainability — ensuring that modernization investments reach the programs and people they are intended to serve, not just the infrastructure that supports them.
Focus should be on treating AI cost, usage, and value as part of the health AI control plane. Agencies should instrument AI platforms from the start so leaders can see which programs are using AI, what it costs, what value it produces, and whether scarce modernization dollars are improving outcomes for beneficiaries, providers, investigators, reviewers, and front-line staff.
Trust and Adoption Have to Be Built Into the Operating Model — For Everyone in the System
AI will not scale if people do not trust the data, outputs, systems, or governance behind it. And that trust challenge runs in two directions: toward the beneficiaries and members of the public these systems serve, and toward the workforce that operates them.
For the generative AI tools increasingly embedded in these workflows, that public-trust standard is now codified, not aspirational. OMB Memorandum M-26-04, implementing Executive Order 14319, requires agencies to procure only large language models that meet two “Unbiased AI Principles” — truth-seeking and ideological neutrality — backed by vendor-level transparency and disclosure requirements. That is the current federal standard for public trust in AI, and it is a more precise standard than “bias monitoring” alone conveys.
This point came through clearly in the summit’s CMS modernization discussions. Leaders emphasized interoperability, shared services, reduced duplication, stronger data foundations, and mission-aligned modernization. The “One CMS” concept reflected a broader shift toward enterprise thinking, where modernization, interoperability, and operational delivery are coordinated across organizational boundaries rather than managed as isolated systems.
But the workforce and change management dimension was equally prominent and urgent. Modernization that outpaces institutional knowledge creates fragility, not capability. Experienced public servants, the ones who understand policy nuance, operational edge cases, historical context, and what happens when a decision goes wrong, are not obstacles to AI adoption. They are essential to making it work. Their knowledge should shape how AI-enabled workflows are designed, tested, governed, and improved over time. Change management is not a downstream activity. It is part of the architecture.
That means “human in the loop” cannot be a checkbox. In high-volume federal environments, people cannot meaningfully review every automated step in every workflow. The better goal is to place public servants where judgment matters most: setting direction, overseeing risk, resolving complex cases, defining system-level outcomes and their validation/evaluation mechanisms, managing exceptions, and making high-consequence decisions. Supporting that workforce is what sustainable AI adoption looks like.
The same logic applies to beneficiaries and end users. Coverage decisions, eligibility determinations, and service interactions carry real weight for the people on the receiving end. AI systems deployed in those contexts need to be accurate, explainable, and subject to meaningful appeal and oversight. Public trust in government services depends on it.
The key is to design AI adoption as an operating model change, not a tool rollout. Federal health agencies should pair technical deployment with workforce enablement, role redesign, end-user feedback loops, appeal pathways, and transparent communications that help both employees and the public understand where AI is used, what it does, and how human judgment remains accountable.
Acquisition Has to Reflect the Accountability Requirement
The summit reinforced that AI-enabled modernization is not only a technology challenge. It is a people, process, acquisition, and governance challenge — and acquisition is where much of the accountability structure gets built.
HHS acquisition leaders framed acquisition as a strategic mission enabler tied to public health outcomes, operational efficiency, emergency response, accountability, and taxpayer value. Figures shared at the summit put the scale in context: $28.2 billion in FY2025 contract spend, 4,459 active contracts, 56,080 contract actions, 10 contracting activities, 11,024 acquisition personnel, and more than 20 acquisition-related systems.
For AI, this means agencies cannot buy responsible outcomes through technology alone. They need acquisition approaches that define the mission problem clearly — including who is affected and how — clarify performance expectations, establish governance requirements, incorporate workforce readiness and change management obligations, and support long-term sustainability. Acquisition is not a downstream afterthought to governance — GSA is working through exactly this tension in real time. A proposed GSA-wide AI acquisition clause governing data rights, vendor transparency, and safeguarding obligations is under active revision after substantial industry pushback, with comments due August 3, 2026, and a public listening session held July 14, 2026. This is where the accountability debate is actually being negotiated right now, not a settled future requirement.
That connection is harder to build into acquisition vehicles than it sounds. Defining meaningful AI performance metrics requires clarity on the mission problem before the contract is written — earlier in the process than most agencies currently engage their technology partners. Building that discipline into how agencies buy is the work ahead.
For federal health, the envisioned end state is one in which contracts require vendors to define not only technical performance but also health mission value, data rights, transparency, governance controls, monitoring obligations, workforce adoption support, and evidence of sustained value over time.
The Next Phase of Federal Health AI
Federal health agencies are operating at the intersection of rising mission demand, workforce strain, technical debt, cybersecurity risk, acquisition complexity, and growing expectations for faster, more responsive digital services. The people those agencies serve — beneficiaries, patients, providers, caregivers — are waiting on the other side of those systems.
AI can help agencies meet that moment. But only when it is governed at the level of responsibility the mission requires, adopted with the change management discipline the workforce deserves, and designed with the full picture in view: the analysts and investigators who use it, the program officers who depend on it, the beneficiaries affected by it, and the public whose trust it must maintain.
The agencies that lead will not be the ones that move fastest without structure. They will be the ones building deliberately toward greater precision about what AI is doing, where risk increases, what controls are required, and how outcomes are measured for everyone in the system. That work is genuinely hard. Federal budget structures, procurement timelines, legacy architecture, and workforce constraints make precision governance difficult to stand up and harder to sustain. That is not an argument against it. It is an argument for building toward it intentionally, incrementally, and with partners who understand both the aspiration and the operational reality.
Federal health agencies do not need AI experimentation at the expense of trust, cost control, mission continuity, or the people these programs exist to serve. They need disciplined modernization that connects use cases to governance, data readiness, workforce adoption, change management, acquisition strategy, and measurable outcomes for real people.
The opportunity is real. So is the complexity. The agencies and partners willing to engage both with discipline, clarity, and genuine accountability to the people at the center of these missions are the ones positioned to lead.
Updated July 2026: Data and figures cited reflect information shared at the 2026 AFCEA Bethesda Health IT Summit and publicly available agency sources. CMS fraud-prevention figures reflect public reporting by Federal News Network on remarks from Bethany Messick, acting deputy director of the CMS Center for Program Integrity, and by Nextgov/FCW on remarks from CMS Deputy Administrator and Chief Operating Officer Kim Brandt.
