Skip to content
Article reader Listen + reading controls
LISTEN + READ YOUR WAY

Article reader

Preparing the reader…

0:00 0:00
Reading settings
Text size
100%

The CDAO evaluation was really about decision rights

When the Pentagon announced that its inspector general was evaluating the Chief Digital and Artificial Intelligence Office in early 2024, it was tempting to ask for a verdict: Was the new organization working?

The final report produced a more revealing answer. The Department of Defense (DoD) Office of Inspector General (OIG) found that the Chief Digital and Artificial Intelligence Office (CDAO) was still operating without several foundational documents needed to make its responsibilities clear across the Department.

The problem was not a lack of ambition. CDAO had inherited data, analytics, digital services, and artificial intelligence (AI) responsibilities from four organizations. It was expected to set strategy, create policy, break adoption barriers, provide enabling services, and scale proven capabilities across one of the world's largest enterprises.

The problem was that a broad mandate does not tell thousands of people how authority should work at the boundary between organizations. Enterprise AI offices succeed or fail in those boundaries.

Revised and substantially expanded July 17, 2026, using the completed OIG evaluation, subsequent CDAO charter, and later organizational realignment.

What the evaluation found

The OIG released Report DODIG-2025-039 on November 14, 2024. It found that CDAO was developing an implementation plan for the Department's 2023 Data, Analytics, and AI Adoption Strategy and working on a chartering directive and implementing instruction. As of June 2024, however, those documents remained in draft.

The delay had practical consequences. CDAO and the DoD Chief Information Officer (CIO) disagreed about parts of the boundary around digital infrastructure, information technology, cloud systems, and related responsibilities. Both offices were concerned about overlap. Without a formal charter and implementation structure, components could not reliably determine which organization owned a decision or service.

The OIG recommended that CDAO publish an implementation plan with strategic performance measures and complete the directive and instruction defining its role. By the time the report was finalized, CDAO had provided an approved implementation plan, closing one recommendation. The remaining recommendation was resolved but open pending publication and verification of the policy documents.

Four days after the report date, the Department issued DoD Directive 5105.89, formally chartering the CDAO. The directive clarified its roles as the principal advisor for DoD adoption and integration of data, analytics, and AI; the Department's Chief AI Officer and Chief Data Officer; and an organization responsible for enabling digital services and enterprise frameworks.

That chronology matters. Oversight did not merely grade past activity. It exposed an operating ambiguity while the institution was still being built.

Mandates do not allocate decisions

An enterprise office can be “responsible for AI” without having the authority, capacity, or proximity to make every decision that phrase implies.

Data ownership sits with mission and functional communities. Cybersecurity and information infrastructure involve CIO authorities. Acquisition decisions move through program and contracting organizations. Testing, legal review, operational risk, intelligence, personnel, and financial management each have their own accountable officials. Military departments and combatant commands operate in different contexts.

The central question is not whether CDAO or a component owns AI. It is which decisions should be centralized, which should remain distributed, and how conflicts move between them.

Central ownership is useful for decisions that create enterprise leverage:

  • common data, platform, and evaluation services;
  • shared technical and evidence standards;
  • portfolio visibility and reusable acquisition patterns;
  • enterprise agreements and cross-component interoperability;
  • authoritative policy interpretations;
  • common talent and training mechanisms; and
  • escalation of barriers no component can remove alone.

Distributed ownership is necessary for mission outcomes, local workflow, operational acceptance, domain data meaning, user adoption, and many risk decisions. A central office cannot understand every context closely enough to substitute for accountable mission leadership.

The operating model needs both. Centralization should remove repeated friction and fragmentation. It should not remove responsibility from the people who field and use the capability.

Define interfaces, not just boxes

An organization chart shows reporting relationships. It rarely tells a program manager what to do when a cloud platform falls under CIO policy, a model uses CDAO data services, a component owns the mission, and an acquisition organization controls the contract.

Useful governance defines the interface:

  1. Trigger. Which event requires coordination—a new acquisition, data source, deployment, material model change, incident, or enterprise-scale request?
  2. Inputs. Which evidence and decisions must the program bring?
  3. Authority. Who recommends, who concurs, who decides, and who can accept an exception?
  4. Time. How quickly must the interface operate at experiment, acquisition, and operational tempo?
  5. Escalation. Where does a disagreement go, and who resolves competing enterprise and mission priorities?
  6. Record. Where is the decision and rationale preserved for later teams and oversight?

Without those details, coordination becomes a meeting. With them, it becomes an operating mechanism.

The OIG report's discussion of CDAO–CIO ambiguity is a case study in why this matters. “Digital infrastructure” can reasonably be understood through both information-enterprise and AI-delivery lenses. Writing a sharper definition helps. A working interface for investment, architecture, security, service ownership, and dispute resolution helps more.

Measure adoption as an enterprise outcome

The 2023 adoption strategy explicitly focused on the organizational environment for continuous delivery, not on identifying a few winning AI applications. It called for interoperable infrastructure, an empowered workforce, advanced research, quality data, governance, and integration with allies and partners.

Those ambitions require measures that cross organizational boundaries.

Counting models, contracts, pilots, marketplace listings, or platform users can show activity. Enterprise effectiveness is better reflected in questions such as:

  • How long does a mission team take to move from a validated need to representative evaluation?
  • How often can teams reuse data, platforms, controls, contract terms, and test assets?
  • Can software and model changes reach users safely at operational tempo?
  • Which barriers recur across components, and how quickly are they removed?
  • Do fielded capabilities improve decision quality, speed, resilience, or cost relative to a baseline?
  • Can the Department observe model, data, vendor, and infrastructure dependencies?
  • Are failed experiments preserved as learning and stopped before they become permanent programs?
  • Do components know where to go for a decision, and do they receive one in time?

The implementation plan requested by the OIG was important because strategy becomes governable only when outcomes, owners, and measures connect.

Reorganization is another test of the interfaces

In August 2025, the Department realigned CDAO under the Under Secretary of Defense for Research and Engineering (USD(R&E)). The stated purpose was to connect research, engineering, experimentation, and AI adoption more closely.

The move may strengthen technical transition. It also changes organizational interfaces. CDAO still has to work across acquisition, CIO, operational, policy, intelligence, test, and military-department authorities. A new reporting line does not dissolve those dependencies.

Every realignment should therefore be accompanied by an interface review:

  • Which decisions moved with the office?
  • Which budget and service responsibilities changed?
  • Which councils, working groups, and escalation paths remain authoritative?
  • What should components do differently on Monday morning?
  • Which measures will show whether the change accelerated mission delivery rather than only moving boxes?

Reorganization creates value when it reduces the cost of coordination and clarifies accountability. Otherwise, it transfers ambiguity to a new part of the chart.

What an enterprise AI office should provide

The CDAO experience offers a practical test for similar offices in government and industry.

An enterprise AI organization should make it easier for a mission team to:

  • choose an appropriate technical path;
  • find and use authoritative data;
  • access secure, observable delivery infrastructure;
  • understand applicable policy and evidence requirements;
  • acquire technology without surrendering essential rights or visibility;
  • evaluate performance in representative conditions;
  • reuse work from other teams;
  • identify who can make a timely risk decision; and
  • scale, constrain, or stop a capability based on evidence.

If the central office adds another approval layer without providing those capabilities, it is governing scarcity. If it supplies platforms without clear ownership and operating rules, it is creating technical dependency. Its product is the combined system: services, standards, decision rights, knowledge, and escalation.

Effectiveness lives between organizations

The 2024 evaluation did not show that CDAO had failed. It showed that consolidating several organizations and publishing an ambitious strategy were insufficient to establish an enterprise operating model.

The later charter addressed important formal gaps. The real test remains behavioral. Can components tell who decides? Can teams reach reusable services? Can disagreements be resolved without months of coordination? Does evidence move from experiments into portfolio decisions? Do fielded systems improve mission outcomes while remaining observable and governable?

An enterprise AI office should be judged less by how much territory its mandate covers than by how effectively the institution moves across that territory.

That is why the OIG evaluation was ultimately about decision rights. The hard part of AI adoption is not assigning one office responsibility for the future. It is designing the relationships through which thousands of people can build that future together.

Sources

READER-NEUTRAL SUBSCRIPTION

Follow Field Notes via RSS.

Copy this address into the RSS reader you already use. New notes will appear there automatically—no account, email address, or tracking required.

Open the feed