The race for AI supremacy will not be won in the lab. It will be won in the courtroom, the treaty hall, and the export-control office — and right now, the legal infrastructure is losing.
That is the central argument of a growing chorus of national-security analysts, lawmakers, and legal scholars who say the United States must pivot from ethics-based AI governance to a deterrence-based model, one that imposes real costs on bad actors rather than relying on voluntary commitments and aspirational principles. The shift is already underway, but the gap between the speed of AI development and the speed of legal adaptation continues to widen — creating vulnerabilities that adversaries are eager to exploit.
"The real AI contest will hinge less on who builds faster chips than on which legal systems can adapt and coordinate the fastest," argued a Washington Times analysis published April 29, framing the competition as one that will play out through export controls, investment reviews, and evidentiary rules rather than through raw computational power alone. The piece captures a mood that has been building across Washington for months: the realization that technology policy without legal teeth is just a press release.
The Deterrence Pivot
For years, the dominant approach to AI governance in the United States was rooted in ethical frameworks — principles of fairness, transparency, and accountability that companies were encouraged, but rarely compelled, to follow. The National Institute of Standards and Technology's AI Risk Management Framework, released in its original form in 2023, exemplified this approach: comprehensive, thoughtful, and entirely voluntary.
That era is ending. The White House's National Policy Framework for Artificial Intelligence, published in March 2026, represents the clearest signal yet that the administration views AI governance through a security lens first and an ethics lens second. The framework proposes federal preemption of state AI laws — an attempt to create a unified legal architecture that can move at the speed the threat demands — though Congress has repeatedly stripped preemption provisions from pending legislation, leaving the regulatory landscape fractured across more than 40 states with active AI bills.
The EU has taken a more prescriptive path. The EU AI Act, which becomes fully applicable on August 2, 2026, imposes a tiered penalty structure that reaches up to 35 million euros or seven percent of global turnover for serious violations. Whatever its flaws, the Act represents something the U.S. still lacks: a unified statute with enforceable consequences.
"Only three percent of compliance professionals report preparedness for incoming AI regulations," according to a 2026 survey cited by Governance Intelligence, a figure that underscores how far the private sector remains from operationalizing any governance framework, voluntary or mandatory.
Export Controls as the New Front Line
Nowhere is the deterrence model more visible than in semiconductor export controls, which have become the sharpest legal instrument in the AI competition. By constraining access to training-class graphics processing units and advanced chip-packaging technology, the United States has attempted to increase the time and capital required for adversaries to replicate frontier AI capabilities — making acquisition uneconomical and slow rather than merely illegal.
The strategy has four core objectives: limiting China's access to American AI chips, blocking access to design software, halting domestic Chinese manufacturing capacity by curbing equipment exports, and restricting technologies that support high-end AI development infrastructure. The global semiconductor market is projected to reach approximately 975 billion dollars in 2026, approaching the one-trillion-dollar milestone, and the chips that train large language models sit at the center of that growth.
But the chokepoint strategy is leaking. Banned NVIDIA GPUs — including the A100 and H100 — have been documented in Chinese university research labs and startup filings, highlighting the difficulty of enforcing chip-level trade restrictions across complex global supply chains. A grey market is growing faster than the regulatory apparatus designed to contain it.
The vulnerability was underscored in 2025 when three engineers at TSMC, the Taiwan-based foundry that manufactures the most advanced chips in the world, were charged with leaking data on the company's 2-nanometer process technology — capable of cutting AI power consumption by roughly 20 percent. The case moved economic espionage from an isolated concern to a systemic risk within the semiconductor supply chain.
States Step In Where Congress Stalls
While Congress continues to debate — and delay — comprehensive federal AI legislation, state legislatures are filling the vacuum. Pennsylvania offers a case study. The state Senate voted 49 to 1 to approve Senate Bill 1090, the Safeguarding Adolescents from Exploitative Chatbots and Harmful AI Technology Act, known as the SAFECHAT Act. Sponsored by Senators Tracy Pennycuick and Nick Miller — the majority and minority chairs of the Senate Communications and Technology Committee — the bill requires companies operating companionship-focused chatbots to disclose the AI's nonhuman status and establish protocols to prevent content related to suicide or self-harm.
At least three additional AI public safety bills are advancing through Pennsylvania's House committees, including HB 2006 on companion chatbot safeguards, HB 2100 regulating mental health chatbots used by therapists, and HB 1857 requiring businesses to disclose AI use in consumer interactions. "This is a common sense step toward protection," Pennycuick said of the SAFECHAT Act, a sentiment that reflects the pragmatic, safety-first approach states are taking in the absence of federal action.
The pattern is not unique to Pennsylvania. More than 40 states have introduced AI-related legislation in 2026, creating a patchwork of rules that some industry leaders warn will be more burdensome than a single federal standard — and that the White House's preemption proposal was designed to prevent.
The Coordination Challenge
The deepest governance gap may be the one between nations. The United States, the European Union, and China are each building distinct legal architectures around AI — the U.S. favoring market-led innovation with targeted controls, the EU treating AI and semiconductors as strategic public assets governed by statute, and China deploying an expanded counterespionage law with bounties targeting Taiwanese semiconductor personnel and export controls on rare earths and high-end magnetics as coercive legal leverage.
"The assumption that chips are a permanent chokepoint has already been undermined by algorithmic adaptation, enforcement gaps in export controls, and a grey market that is growing faster than the regulatory apparatus designed to contain it," warned researchers at Chatham House in an April 2026 analysis that questioned whether export controls alone can sustain the deterrence model.
What to Watch
Three developments will determine whether the shift from ethics to deterrence produces a coherent governance regime or simply a harder-edged version of the same fragmentation. First, whether Congress can pass any federal AI legislation before the EU AI Act's August 2026 enforcement deadline effectively sets the global standard by default. Second, whether semiconductor export controls can be tightened and enforced fast enough to stay ahead of grey-market circumvention and algorithmic workarounds that reduce dependence on frontier hardware. And third, whether the growing patchwork of state laws — from Pennsylvania's chatbot safety bills to California's AI transparency requirements — forces the federal government's hand on preemption, or whether the regulatory fragmentation becomes the permanent American answer to a problem that demands unity.
The AI security contest is no longer about who builds the smartest model. It is about who builds the legal system that can keep up.
"The real AI contest will hinge less on who builds faster chips than on which legal systems can adapt and coordinate the fastest."— Washington Times, Analysis