From governance to execution in federal AI policy

Global artificial intelligence (AI) spending is forecasted to reach $2.5 trillion in 2026, according to one estimate, marking a 44% increase year-over-year. In 2027, spending is expected to reach $3.3 trillion, reflecting another 32% increase year-over-year. The majority of this money has funded AI infrastructure, followed by AI services and AI software (representing about 54%, 23%, and 18% of spending, respectively, for 2026).

As discussed in our most recent Brookings report, federal AI spending in the U.S. is also rapidly increasing and at a rate faster than the rest of the world. Our research showed that the value of funds obligated increased by 966% between 2024 and 2026 (from $355 million to $7.2 billion), and the value of potential award increased by 1,912% for the same period (from $4.6 billion to $91.8 billion).

Out of 441 total federal agencies, the number of agencies with AI contracts rose from 17 in 2022 to 23 in 2024 and 28 in 2026. While the U.S. Department of Defense accounted for by far the most AI contract spending (at $90 billion), other agencies with AI contracts include the Department of Commerce ($197 million), the Department of Health and Human Services ($138 million), and NASA ($45 million). Even small agencies—such as the National Archives and Records Administration ($110,000), the Nuclear Regulatory Commission ($890,000), and the Department of the Interior ($1.5 million)—had contracts, reflecting how AI is spreading across the federal government.

As the number of agencies with AI contracts rose, so did the total number of contracts across the federal government—from 472 in 2022 to 961 in 2024 and 1,743 in 2026. While many of these contracts across all years were categorized under “Professional, Scientific, and Technical Services,” others were in construction, manufacturing, information and cultural industries, and educational services.

Anticipating this growth, the Biden administration created the Chief AI Officers Council (CAIOC) in 2024 to shift guidance for AI from agency-by-agency initiatives to interagency coordination. As designed, the CAIOC sits at the intersection of strategy and execution for AI initiatives across the federal government and offers insights into how federal AI is evolving. Examining the role and composition of the CAIOC under Biden reveals how his administration approached AI governance compared with the current Trump administration.

Drawing on Leadership Connect data on CAIOC membership, we identified the 17 members of the Biden-era council and nine members of the current Trump-era council.¹ We collated information on each member from Leadership Connect data and supplemented it with data from LinkedIn and public-facing government websites to examine career profiles, qualifications, and personal characteristics. Biden-era members are coded according to their most recent role prior to CAIOC membership while Trump-era members are coded according to their current role.

Governance models

The Biden and Trump administrations’ executive orders and Office of Management and Budget (OMB) memos on federal AI reveal starkly different approaches. The two administrations do not merely differ in rhetoric and ideology but also structured federal AI governance around two very different primary purposes (see Table 1).

The Biden administration’s approach could be termed “governance-led adoption,” reflecting the advancement of AI through structured governance, risk management, use-case inventories, and minimum safeguards for uses affecting rights and safety. In other words, the Biden administration saw AI’s potential but also sought to manage its potential risks.

By contrast, the Trump administration’s approach could be termed “adoption-led governance,” in which AI use is accelerated across government agencies. While governance frameworks are retained, they are repositioned to remove friction, speed up deployment, and support U.S. competitiveness and “global AI dominance,” particularly against China. The Trump administration has acknowledged potential risks, as evidenced by its AI Action Plan, but remains far more focused on accelerating AI deployment to achieve global dominance than regulating it.

While the form of the council remained the same across both administrations, the governing logic shifted from responsible-use coordination to implementation-focused coordination. This interpretation is supported by our examination of federal AI spending under both administrations. Further, this shift in approach is not only reflected in the formal policy language, but also in the backgrounds of officials selected to serve on the CAIOC.

Personnel

A president’s administrative priorities are expressed not only through formal rules but through the backgrounds of the people selected to implement them. Senior leaders bring different experiences and habits of thought, and the CAIOC inherits the priorities and perspectives of those who lead it. These changes in composition can affect coordination style, risk tolerance, implementation choices, and interagency relationships. With that in mind, we analyze the backgrounds of CAIOC leaders under both administrations (see Table 2).

In terms of composition, the Biden-era council was broader, more diverse, and more cross-sector, while the Trump-era council is more operationally concentrated and more heavily weighted toward security, cyber, and enterprise execution, particularly in prior government positions. For example, 89% of Trump council members had their previous job in government, compared with only 19% of Biden council members. This suggests the Biden council was heavily focused on generating ideas about how to use AI, while the Trump council is more focused on applying AI within a governmental setting. That shift is consistent with a broader move in federal AI spending from strategic planning to more operational contracts.

The change in the CAIOC across administrations matters not only because it reflects different priorities but because it likely shapes how federal AI governance works in practice, particularly given varying levels of expertise and focus among members. Although interagency councils like the CAIOC do not directly implement policy, they influence how agencies coordinate and what issues they address. In that sense, the council’s changing composition offers a useful indicator of the federal government’s preferred operating model for AI.

The respective councils are likely to coordinate differently

Under the Biden administration, both the formal policy framework and the council’s composition suggested a body designed for broad governance coordination. The wider mix of policy, science, health, budget, legal, and digital modernization backgrounds fits a council expected to balance multiple goals at once: innovation, risk management, public trust, administrative consistency, and cross-agency oversight. In practical terms, that kind of body is more likely to emphasize common process, shared governance language, inventories, review mechanisms, and structured coordination across agencies with very different missions—a natural approach in the more nascent era of federal AI adoption.

Under the Trump administration, the council’s composition is more concentrated in cyber, CIO, data, security, and operational leadership, as well as government applications. That suggests a different coordinating style: one less focused on building broad governance consensus and more concerned with identifying operational barriers, accelerating implementation, and spreading usable practices across agencies. This is consistent with Trump’s AI Action Plan, which is heavily implementation-focused. That shift is also reflected in the federal government’s current spending priorities.

Agency AI leadership may be judged by different criteria

The shift in orientation likely has implications for how agency chief AI officers (CAIOs) and related technology leaders—chief information, technology, and data officers—are evaluated. In the governance-centered Biden model, CAIOs were valued for their ability to coordinate internal oversight, build governance processes, align with OMB requirements, and manage institutional legitimacy. In the implementation-centered Trump model, CAIOs are more likely to be valued for delivery—moving systems into operation, supporting mission use cases, overcoming bureaucratic bottlenecks, and integrating AI into enterprise workflows. Success becomes a question of not only whether AI is appropriately governed but whether it is being deployed in ways that improve performance and demonstrate tangible value. Given that the Trump council is much more heavily weighted toward former government employees, they were likely chosen for demonstrated success in government.

This shift is likely reflected in incoming CAIOs and may favor officials with stronger operational, cyber, data, and systems backgrounds over those with comparative strength in governance design, cross-sector stakeholder management, or policy coordination. It may also elevate the internal status of CAIO roles within agencies, bringing them closer to the mission delivery functions of chief information, technology, and data officers.

Risk management does not disappear, but it changes function

One of the most important implications of the shift in CAIOC composition involves risk management. Under the Biden administration, risk management was central to the architecture of AI governance. AI could be useful but only if agencies had clear structures for oversight, review, documentation, and public accountability. In that model, governance often operated as a control mechanism. Under the Trump administration, risk management has not disappeared, but it functions more as a supporting guardrail within a broader adoption agenda. It is no longer a key organizing principle but one consideration among several in a model oriented toward deployment, competitiveness, and execution. In short, under Trump, successful implementation is the primary yardstick by which CAIOs are measured.

This matters because it changes the relative authority of legal, policy, and oversight officials. In a governance-led model, review and accountability functions carry stronger institutional leverage. In an implementation-led model, those same functions may be expected to facilitate deployment rather than set the pace of it. The practical result is likely less emphasis on process-heavy assurance and more emphasis on operational mitigation—identifying risks that truly block deployment while moving the rest of the system forward.

The shift from governance to operational orientation carries both benefits and risks. A governance-oriented CAIOC tends to support standardization across agencies. In the Biden administration, it created common language, common expectations, and a stronger presumption that agencies should move in parallel. The Biden council was more focused on bringing big ideas in from outside government than on their implementation.

Today’s more operations-oriented CAIOC may instead support differentiated implementation, where agencies move at different speeds depending on mission need, technical capacity, and leadership readiness. That approach may produce faster uptake in agencies with clearer mission applications, stronger technical staff, or more mature data environments—as seen in the strong concentration of AI spending at the U.S. Department of Defense.

However, it also creates risks. If the center of gravity moves toward implementation, differences in agency capacity may become more pronounced—already visible in the disparity between Defense Department spending and that of the other 27 agencies (out of 441 total agencies) with AI contracts. Some agencies may move quickly and effectively while others lag or adopt unevenly. In that environment, the CAIOC’s role becomes especially important not just to coordinate policy but to reduce capacity gaps by spreading practices, talent, and implementation models across agencies. Federal AI governance may increasingly be experienced not as a uniform rules regime but as a contest over execution capacity between agencies.

The Trump-era CAIOC is still in the early stages of organization but is expected to move quickly in shaping federal AI spending and execution. Given the explosive growth in federal AI spending and the broader expansion of AI globally, the council faces significant work ahead.