The White House’s AI Action Plan
TL;DR
Background: In October 2023, former President Joe Biden signed an executive order on artificial intelligence that emphasized caution, with a focus on governance frameworks, safety protocols, and AI risk management. The Trump administration’s just-released AI Action Plan, in contrast, treats the development of AI as a national imperative akin to the Space Race, with a coordinated plan to advance innovation, infrastructure, and international leverage.
And… The plan calls for aggressively removing state and federal restrictions that could slow AI progress, repurposing federal lands for data centers and energy infrastructure, and channeling federal resources toward chip manufacturing, open-source models, and regional AI hubs. It also proposes denying federal funds to states that impose burdensome AI rules. Its animating assumption is that AI infrastructure and model development are essential to maintain U.S. economic and geopolitical dominance.
However… The plan is silent on copyright, which has been one of the most hotly contested legal questions facing generative AI. The omission may reflect limits on executive authority. But given the U.S. Copyright Office’s recent criticism of fair use and ongoing litigation over model training, this silence is conspicuous. The absence of any position leaves unresolved a core risk to AI deployment: the threat of unpredictable copyright liability for foundational model developers.
KEY TAKEAWAYS
A Full-Stack Strategy
A centerpiece of the AI Action Plan is its embrace of a “full-stack” strategy that treats AI not merely as a set of models, but a vertically integrated ecosystem that spans research, infrastructure, talent, and deployment. Moreover, the federal government would play a coordinating role across each layer of this stack. It proposes new federal investments in scientific labs equipped for AI-accelerated experimentation, as well as long-term support for focused research organizations aimed at enabling breakthroughs in material science, biology, and engineering. These initiatives are coupled with a forthcoming “National AI R&D Strategic Plan,” which is intended to guide federal priorities toward frontier scientific questions, including interpretability, robustness, and model control.
The plan identifies energy and computing-power constraints as central bottlenecks to the future development of AI to address these barriers, it calls for permitting reform to facilitate rapid construction of data centers and semiconductor manufacturing facilities, particularly on federal lands. To address shortfalls in access to computing power, it proposes creating spot and forward markets for cloud access.
Notably, the plan gives special attention to open-weight and open-source models, framing them as geopolitical assets. These models, it argues, should be accessible to researchers and smaller firms that cannot afford the proprietary licenses and data restrictions of closed models. The federal government would support these efforts by granting expanded access to the National AI Research Resource (NAIRR), new partnerships with private cloud providers, and coordinated technical support from agencies like the National Institute of Standards and Technology (NIST).
A Geopolitical Framing
While positioning AI technology as a domain of strategic competition, the plan asserts that global leadership in AI requires control over both upstream inputs and downstream influence. Chief among its priorities is continued restrictions on trade of advanced semiconductors and AI computing power from adversarial states like China. The document proposes stronger export enforcement, including deployment of location-verification tools embedded in AI-related hardware to track illicit diversion. These controls would be designed to prevent foreign military and intelligence entities from leveraging U.S.-origin technologies to undermine American interests.
The plan also calls for coordinating export-control regimes with allied nations, signaling a shift toward plurilateral enforcement. It suggests that the United States should pressure partners to adopt complementary restrictions, including through the use of tools like the Foreign Direct Product Rule and secondary tariffs. It further suggests leveraging U.S. dominance in key supply-chain components like lithography, advanced substrates, and software frameworks to prevent backfilling by third parties.
In order to deploy AI exports as a positive instrument of U.S. foreign policy, the plan recommends creating full-stack AI export packages of models, infrastructure, training, and standards for use by allied and partner nations. This would mirror Cold War-era strategies around nuclear energy and telecommunications, whose export was as much about setting global norms as capturing markets.
Federal Preemption by Other Means
While the AI Action Plan does not explicitly invoke federal preemption, it proposes conditioning federal funds on states’ regulatory choices, and withholding discretionary funds from states whose AI laws are overly burdensome or obstructive. This would be similar to earlier efforts to use fiscal leverage to shape states’ behavior. The strategy emerges in response to a growing state-level patchwork of AI rules across the states. The plan treats such heterogeneity not as benign federalism, but as a potential structural impediment to national innovation capacity.
This framing reflects a recognition that overlapping and inconsistent compliance regimes can generate fixed costs that scale poorly and deter new entrants. A temporary federal moratorium on new state AI regulations would allow space for a coherent national framework to emerge that aligns risk-based oversight with the technical realities of AI systems. While the plan stops short of such a moratorium, it acknowledges that the economic consequences of legal fragmentation may rival those of technological stagnation.
The Copyright Question
One of the most conspicuous omissions from the AI Action Plan is the question of copyright—an area of mounting legal uncertainty for developers of generative-AI services. Indeed, the plan says nothing about how foundational models interact with the Copyright Act or how developers should navigate fair use doctrine.
This silence is particularly notable given the Copyright Office’s recent report casting doubt on whether training models on copyrighted material constitutes transformative use. The report implied that model weights might themselves be treated as a form of infringing database or derivative work.
This view risks collapsing the distinction between functional expression and expressive content, misunderstanding how generative models encode statistical relationships, rather than expressive copies. The result is a growing threat that litigation or regulatory overreach—rather than market failure—will dictate the boundaries of lawful AI development. By declining to engage with these questions, the plan effectively punts a critical issue that could undermine the very innovation it seeks to promote.
For more on this issue, see Kristian Stout’s earlier tl;dr “State Approaches to AI Regulation Are a Patchwork”; his WLF Legal Pulse post “Federal Preemption and AI Regulation: A Law and Economics Case for Strategic Forbearance”; and his Truth on the Market post “Bartz v. Anthropic: Mapping Fair-Use Boundaries in the Age of Generative AI.”