We are witnessing the unfolding of a regulatory regime for artificial intelligence that mirrors, in many of its features, the federal structure of the American republic. Just as the Framers divided sovereignty between the general government and the states, we now find authority over AI innovation and safety distributed across multiple jurisdictions—executive orders at the federal level, legislative experiments in the several states, and binding treaties abroad. The genius of any well‑constructed framework lies in its capacity to balance energy in the executive with the sober deliberation of local bodies; yet the present landscape exhibits a dangerous tendency toward both concentrated federal power and a disunited patchwork that threatens the very innovation it purports to nurture.
At the federal level, the absence of a comprehensive AI law has left the field to executive action and agency guidance, creating a bifurcated approach that is at once permissive and coercive 62,77. The recent “Promoting Advanced Artificial Intelligence Innovation and Security” executive order of June 2026 establishes a voluntary pre‑release review process for frontier models 7,13,66, explicitly eschewing mandatory licensing or preclearance 72,73 and signaling a philosophy of light‑touch oversight 29. Yet this same order retains the authority to suspend model access on national security grounds 63 and directs the Department of Justice to challenge state AI laws 65 under a preemption rationale, so as to prevent what is styled a “patchwork” of regulation 28,57. This is a dual posture that introduces considerable regulatory uncertainty: an ostensibly voluntary framework backed by potent intervention tools, as demonstrated by the 19‑day global suspension of Anthropic’s Claude models under an obscure export control directive 63,101,102,103 and the constrained rollout of OpenAI’s GPT‑5.6 56,66,90. For a firm like Alphabet, which operates frontier models such as Gemini, such abrupt shifts disrupt product launches and undermine long‑range planning 46.
The great danger here is the accumulation of unchecked authority in the executive branch, exercised without the deliberate consent of Congress. By invoking export controls and national security as blanket justifications, the administration may act in ways that profoundly affect interstate and foreign commerce, yet its actions remain largely insulated from legislative oversight. We must ask: does this allocation of authority create a system of mutual oversight, or does it concentrate power in a manner that invites abuse and unpredictability?
The States as Laboratories—and the Perils of Disunity
The states, in their traditional role as laboratories of democracy, have responded to federal inaction with great vigor: in 2025 alone, 145 AI‑related laws were enacted across the several states 23,77,98. Colorado’s pioneering consumer protection law, though now largely stripped of its core provisions by legal challenges and subsequent amendment 51,65, and frameworks such as Texas’s TRAIGA 2.0 38,77 and Utah’s prohibitions on AI grading 25,26 illustrate the diversity of approaches. Yet the federal push for preemption has intensified. The Federal Trade Commission has proposed that certain state AI laws are impliedly preempted 64, and the President has threatened to withhold funding from states enforcing algorithmic discrimination provisions 65. This clash of powers creates a volatile environment where Alphabet must maintain a dynamic compliance posture across 50 jurisdictions, increasing legal and operational overhead 88,89.
The tension is familiar from our constitutional history: the Commerce Clause, for instance, has long served as a bulwark against state regulations that unduly burden interstate commerce. A well‑crafted preemption doctrine can provide uniformity and reduce compliance costs; yet if it is applied too broadly or wielded as a political instrument, it may stifle local democratic experimentation and leave citizens without recourse against novel harms.
Infrastructure and the Clash of Federal Facilitation with Local Sentiment
The physical infrastructure that undergirds AI—data centers—presents a further jurisdictional conflict. The federal administration actively supports the build‑out of these facilities, with the Federal Energy Regulatory Commission accelerating interconnection 59 and the Energy Secretary urging expedited grid connections 49,60. Yet direct democratic sentiment runs strongly in the opposite direction: polling indicates that 7 in 10 Americans oppose local AI data center construction 12,33,42,55, and communities from Florida counties 76 to Brown County, Wisconsin 47 have considered moratoria. Environmental litigation, such as the NAACP’s action against xAI for air pollution 69, signals emergent risks that may delay or burden new projects.
Here, the constitutional analogy lies in the tension between the federal government’s authority over interstate commerce and the states’ police power to protect health, safety, and welfare. The federal government may facilitate, but it cannot command the assent of local communities; unless a modus vivendi is found, the necessary infrastructure for AI advancement may be caught in a paralysing stalemate.
Export Controls and the Deemed Export Doctrine
Export controls, originally conceived to restrict the transfer of semiconductor technology 75,95, have now been extended directly to AI model access. The Export Control Reform Act has been invoked to temporarily ban foreign national access to Anthropic’s models 70,100 and to delay OpenAI’s release of GPT‑5.6 pending government approval 52. These interventions are cast in the language of national security 74, yet the legal architecture—particularly the “deemed export” doctrine applied to cloud APIs 63—retains a broad and unpredictable sweep. Although some restrictions have been partially lifted 45,79, the underlying authority remains intact, capable of being reactivated at any moment. For Alphabet, which depends on international talent and serves a global customer base, such controls may fracture service offerings, restrict engineer access, and provide foreign competitors with an opening to develop homegrown alternatives 63.
Historical parallels with encryption controls are instructive. Overly restrictive export regimes in the 1990s did not prevent the spread of strong cryptography; instead, they spurred the development of foreign products and eroded American preeminence 14. We risk repeating that error if we apply the same heavy‑handed measures to AI.
The Labour Question and the Social Compact
The displacement of labour by AI is a matter of profound social consequence, yet no federal law currently mandates disclosure of AI’s role in layoffs 78,99. Proposed legislation, such as the AI Workforce PREPARE Act 78,99, and state‑level bills requiring advance notice 77 signal a growing demand for transparency. Public sentiment is ambivalent: a recent survey found that 55% of employers regretted AI‑driven layoffs 82, and companies such as Ford have rehired engineers after a failed automation initiative 44. The administration’s emphasis on AI‑driven economic growth 20 and its $500 million workforce initiative, RaiseUs 91, recognizes the political necessity of addressing labour transitions, but these measures are as yet insufficient.
A well‑constructed framework must balance the economic dynamism that AI enables with the social stability that republican government requires. If the benefits of AI accrue to a narrow segment of capital while broad swaths of the citizenry bear the costs of displacement, the social compact itself may be damaged.
International Fragmentation and the Rise of AI Sovereignty
Beyond our shores, the world is dividing into regulatory blocs. The European Union positions itself as a third pole 68,87, binding firms with the comprehensive AI Act 96, while G7 members advocate “AI sovereignty” to reduce dependence on American technology 54. Canada pursues its “AI for All” strategy 21,32,36,53 and has introduced chatbot regulations 48; Australia debates copyright exemptions for AI training 67,97; and China has imposed strict domestic controls on algorithms and data 15,37 while restricting its AI talent from working abroad 17. Other nations, including Malaysia, Singapore, and Japan, are enacting their own frameworks 18,50,57,61, creating a multipolar landscape that may limit the global scalability of any monolithic, U.S.‑centric AI solution 81.
From a Madisonian perspective, this fragmentation resembles the confederation of sovereigns that preceded our Constitution—a state of affairs where the lack of uniformity and mutual recognition hampers commerce and innovation. A wise policy would seek to establish principles of comity and mutual recognition, rather than insulate American AI behind walls that invite emulation and retaliation.
Governance Frameworks and the Rule of Law
Absent a federal statute, the United States has come to rely heavily on voluntary frameworks such as the NIST AI Risk Management Framework 1,2,3,4,5,6,9,10,11,16,19,22,30,31,34,35,39,41,84,94,96 and ISO/IEC 42001 27,83. These offer valuable guidance but lack the force of law 85. The absence of a consistent federal privacy law 92 compounds compliance complexity, particularly as new proposals seek to restrict AI from reselling consumer health data 43,93. The courts, for their part, have provided important cornerstones: the Supreme Court has denied AI personhood 71 and reaffirmed broad executive removal power 40, while the judiciary increasingly sanctions AI‑generated errors in filings 24. These decisions signal that the courts will not create new legal categories for AI, leaving the field to the political branches—and thereby placing a premium on legislative deliberation.
Implications for Institutional Design and the Private Sector
For a corporation such as Alphabet, which stands at the intersection of these dynamics, the aggregate picture is one of moderate headwinds and strategic complexity. Federal contracts for AI have grown from 472 in 2022 to 1,743 in 2026 86, yet only 28 of 441 federal agencies hold such contracts 86, suggesting vast untapped potential. The administration’s hands‑off posture facilitates rapid innovation, but the looming threat of export controls or model suspensions introduces operational risk that cannot be insured against. The push for federal preemption could, if successful, reduce state‑level compliance burdens; but until the constitutional questions are settled, Alphabet must prepare for a prolonged period of dual jurisdiction.
On the infrastructure front, while the administration supports fossil‑fueled capacity to power AI 58, community and environmental pushback may delay projects and raise costs. Workforce‑related regulations may compel investment in retraining, but such measures are also a necessary investment in the social license to operate. Internationally, Alphabet’s reliance on American infrastructure may become a liability as allies insist on domestic data residency and “trusted partner” designations. The company’s engagement with the AI Safety Institute 80 positions it as a responsible actor, and in a world of fractured trust, such designations carry significant weight 8.
The central question is one of institutional design: How can we allocate authority over AI governance in a manner that preserves the benefits of innovation while checking the forces of consolidation and regulatory arbitrage? The answer cannot lie in a single, all‑powerful federal agency, for that would simply replace one concentrated power with another. Nor can it lie in an uncoordinated multiplicity of state legislatures, each pulling in its own direction. The genius of the Constitution lies in its system of checks and balances—a principle that must now be translated into the domain of artificial intelligence.
A judicious framework would draw clear lines of jurisdiction: the federal government should set minimum safety and security standards and manage international agreements, preserving the supremacy of uniform rules in matters of interstate and foreign commerce. The states, in turn, should retain authority over local impacts—land use, environmental protection, and consumer redress—where their proximity to the citizen yields superior knowledge. Preemption must be a scalpel, not a sledgehammer, applied only where state laws genuinely conflict with federal objectives and not merely when they impose higher standards than the industry desires.
In the end, the greatest risk is not that AI regulation will be too strict or too lenient, but that it will be arbitrary—subject to the sudden decree of an executive or the momentary passion of a statehouse, without the deliberative process that legitimates government action. Until Congress enacts a comprehensive law that vests its authority in a stable institutional design, we shall continue to lurch from crisis to crisis, and the very innovation we seek to protect will be the casualty.