AI Preemption War: Part 2 — The Laboratory States

Series: The AI Preemption War | Table42 Research

—

Before the federal government moved, states were already building the architecture of AI accountability. California, Colorado, Illinois, and a dozen others had spent years developing legal frameworks for algorithmic discrimination, automated decision transparency, and AI-enabled fraud. Their laws weren’t perfect—but they were responsive to local constituents, shaped by state-level debate, and accountable to state voters.

The National Policy Framework’s preemption language threatens to erase that work. Not through a single decisive stroke—through a bureaucratic process of displacement that state officials are only beginning to understand.

What States Built

California led the field. Its AI transparency bills—AB 2013 and SB 942—would require disclosure when AI generates content that consumers rely on, mandatory watermarking of AI-generated material, and vendor accountability for AI-enabled harms. Governor Newsom signed an executive order in March 2026 to strengthen AI vendor safeguards for state contracts. California’s approach reflects a state economy deeply intertwined with AI development—Silicon Valley houses the companies being regulated—and a political culture willing to impose compliance costs on the industry.
Colorado enacted the nation’s first comprehensive AI law in 2024, the Colorado AI Act, targeting algorithmic discrimination in high-stakes AI systems: hiring, housing, credit, healthcare. The law requires impact assessments, human review of automated decisions, and discrimination prevention. Colorado’s law was deliberately modeled on the EU’s approach—tiered risk classification, mandatory safeguards for high-risk systems, enforcement through the state attorney general. It was also delayed, amended, and reworked multiple times as the state navigated industry pushback and technical complexity.
Illinois has the longest track record. Its Biometric Information Privacy Act (BIPA), passed in 2008, predates the current AI debate but has become central to AI regulation because it governs the data AI systems depend on. BIPA requires informed consent before collecting biometric data—fingerprints, facial geometry, retina scans—and imposes substantial penalties for violations. Illinois also passed the AI Video Interview Act (2020), regulating AI analysis of job candidate videos, and the Automated Tool Tax Credit (2024), creating financial incentives for AI deployment that meet state standards.
New York requires bias audits for automated employment decision tools and has established AI task forces to monitor industry practices. Texas created regulatory sandboxes for AI innovation, allowing companies to test AI products without standard regulatory oversight. Washington passed laws governing synthetic media and deepfakes. Maryland imposed limits on facial recognition use by law enforcement.

The result is not a coherent national framework—it’s a patchwork. Some states emphasize consumer protection; others focus on innovation. Some regulate specific AI applications; others set broad principles. But the patchwork reflects something important: states were adapting to AI governance based on their own economic structures, political cultures, and constituent demands.

What Preemption Would Erase

The White House framework’s preemption language targets laws that are “unduly burdensome” to AI development. The term is undefined. What’s clear is that the most detailed state AI laws—California’s transparency requirements, Colorado’s algorithmic discrimination rules, Illinois’ employment disclosure statutes—would likely qualify.

Consider what each state’s laws would face:

California’s AB 2013 and SB 942: These impose affirmative obligations on AI vendors—to disclose AI-generated content, to watermark synthetic media, to maintain audit trails. Under preemption, these requirements could be voided if a company argues they impose compliance costs beyond what federal minimums require. The state’s ability to set standards for AI vendors seeking state contracts would be narrowed.
Colorado’s AI Act: The law’s impact assessment requirements, human review mandates, and discrimination prevention obligations apply to any AI system used in hiring, housing, credit, or healthcare decisions affecting Colorado residents. Preemption would eliminate these protections—or at least create enough legal uncertainty that companies would decline to comply, arguing federal standards preempt the state requirements.
A federal court has now added another layer of uncertainty. On April 28, 2026, a federal judge issued an order preventing Colorado from enforcing SB 24-205 ahead of its June 30 effective date. The ruling came in a lawsuit filed by xAI (Elon Musk’s AI company) arguing the law violates the First Amendment by compelling speech about AI decision-making. The Justice Department joined xAI’s lawsuit, with the administration arguing the Colorado law “jeopardizes the United States’ position as ‘the global AI leader.'” Colorado is attempting to narrow the law’s scope to preserve what can be preserved, but the combination of preemption risk, federal litigation, and DOJ opposition creates multiple pressure points on the law’s viability.
Illinois’ BIPA and AI Video Interview Act: These laws govern how AI companies can collect and process data from Illinois residents. Preemption could narrow their scope, particularly if federal law establishes minimum data protection standards that states cannot exceed.

The pattern is consistent: states built AI laws to protect residents from specific harms they identified. Federal preemption would replace those choices with a national baseline—not because Congress has determined the state laws are unconstitutional, but because the administration has decided they are “burdensome.”

The Resistance

States are not accepting this quietly.

Colorado is reworking its AI law ahead of its delayed 2026 effective date—attempting to align state requirements with the emerging federal framework to preserve what can be preserved. Governor Jared Polis has been publicly critical of preemption that strips state authority without providing meaningful alternatives.
California has ordered new safeguards for AI vendors seeking state contracts, effectively using procurement policy to maintain leverage where legislation may be preempted. The state is also exploring whether preemption challenges can be framed as violations of the federal structure—though constitutional doctrine here is underdeveloped.
A coalition of state attorneys general has begun coordinating on preemption response, sharing legal analysis and potentially joining litigation if a federal AI statute passes with overly broad preemption language. Several have issued statements arguing that preemption without equivalent federal enforcement creates accountability gaps that harm residents.

Congressional resistance exists too. Josh Gottheimer (D-NJ) has warned that preemption without a more comprehensive federal standard will face resistance. Senator Maria Cantwell has signaled openness to a federal standard, but only one “with enough substance to work”—meaning federal protections equivalent to what states built. The House Democratic caucus is holding listening sessions with major caucuses to develop a coordinated response.

The political dynamics are complex. States with large tech sectors—California, Washington, Massachusetts—tend to have both AI companies and AI regulation. States with stronger consumer protection traditions—Illinois, New York, Massachusetts—have more developed AI accountability frameworks. States seeking to attract AI investment—Texas, Florida, Nevada—have been more cautious about regulation. Preemption consolidates power in states that want lighter regulation, at the expense of states that want stronger protections.

The Federalism Question in Practice

What makes the AI preemption fight different from other federalism contests is the pace of technology.

Traditional preemption cases involve laws that have been on the books for decades. Courts have developed doctrines, Congress has drawn lines, industry has adapted. Preemption in those contexts rarely creates sudden disruption—it’s an evolution of a settled legal landscape.

AI preemption would be different. State laws have been enacted in the last one to five years—some are not yet in effect. Companies are still adapting to their requirements. Residents are only beginning to see their effects. Preemption at this stage would erase laws before their impacts are understood, before compliance infrastructure is built, before courts have interpreted their scope.

This matters for the Table42 angle: the AI preemption fight is not just about jurisdiction. It’s about whether democratic governance can adapt to technology at the pace technology requires—or whether tech companies can capture the moment of adaptation to lock in federal preemption that prevents states from ever developing their own frameworks.

The federalism question, in practice, is a question about whose choices govern: state voters and legislators who see AI’s harms up close, or federal officials and industry players who prefer national uniformity.

What the Patchwork Gets Wrong

An honest assessment of state AI regulation must acknowledge that the laboratory has produced its share of failed experiments.

Colorado’s implementation delays. The Colorado AI Act was originally set to take effect in February 2025; it has been delayed multiple times and significantly amended as the state struggled to define compliance requirements, resolve industry objections, and establish enforcement mechanisms. The law’s effective date is now 2026—two years after passage—and the state still lacks final rulemaking guidance. If this is the laboratory, the experiment is running behind schedule.
California’s SB 1047 veto. Governor Newsom vetoed the most ambitious state AI safety bill in September 2024, specifically because it was poorly designed. SB 1047 would have required safety testing and certifications for frontier AI models, but Newsom’s veto message criticized the bill for applying “stringent standards only to the largest and most expensive models” while ignoring smaller, potentially riskier systems. The veto reflected not industry capture but genuine concern about regulatory design—a state acknowledging that its own legislative product was flawed.
BIPA’s litigation burden. Illinois’ Biometric Information Privacy Act has generated thousands of lawsuits—over 1,400 cases filed between 2017 and 2023—many against companies that had no biometric misuse but had procedural compliance failures. The result has been substantial litigation costs without clear evidence that the law has reduced actual biometric data misuse. BIPA demonstrates that even well-intentioned state regulation can create enforcement patterns divorced from its protective purpose.
Technical infeasibility criticisms. Multiple state AI laws have been criticized by technologists for requirements that are difficult to implement. New York’s bias audit requirement for automated employment tools was initially so vague that companies didn’t know what constituted compliance. Colorado’s impact assessment requirements reference technical standards that don’t yet exist. The gap between legislative ambition and technical reality is real.

These problems don’t invalidate state-level AI regulation. But they complicate the narrative that states are reliably producing sound governance that preemption would destroy. Some state laws are well-designed; others are not. A preemption debate that ignores the failures is as incomplete as one that ignores the successes.

Primary Sources

State AI Laws

1. California AB 2013 (Chapter 673, Stats. 2024): Generative AI transparency requirements for developers — signed September 28, 2024, effective January 1, 2026
2. California SB 942 (Chapter 674, Stats. 2024): AI watermarking and disclosure requirements for AI-generated content
3. Colorado SB 24-205 (2024): Colorado Artificial Intelligence Act / Anti-Discrimination in AI Law (ADAI) — comprehensive consumer protection, effective February 1, 2026
4. Illinois Biometric Information Privacy Act (BIPA), 740 ILCS 14/1 et seq. (2008): Consent requirements for biometric data collection — longest-standing state AI-related regulation
5. Illinois AI Video Interview Act, 820 ILCS 42/1 et seq. (2020): Disclosure and consent requirements for AI analysis of job candidate videos
6. Illinois HB 3773 (Public Act 103-0606, 2024): Amends Illinois Human Rights Act — employer notice when AI influences employment decisions
7. New York Local Law 144 (2023): Automated Employment Decision Tool (AEDT) bias audit requirements
8. Texas HB 149 (TRAIGA, 2025): Texas Responsible Artificial Intelligence Governance Act — prohibited uses, disclosure requirements, civil penalties
9. Texas SB 1674 (2023): AI regulatory sandbox for innovation — testing without standard regulatory oversight
10. Washington HB 2031 (2023): Synthetic media and deepfake disclosure requirements
11. Washington SB 5879 (2023): AI task force for monitoring industry practices
12. Maryland HB 276 (2022): Facial recognition use limitations for law enforcement
13. Virginia HB 2034 (2022): Consumer data protection with AI implications
14. Connecticut SB 1103 (2023): AI working group and impact assessment requirements for state agencies
15. Vermont H 510 (2024): AI impact assessment requirements for state-contracted AI systems
16. Executive Order N-12-26 (California, March 2026): Governor Newsom’s AI vendor safeguards for state contracts

Federal Framework and Analysis

17. National Policy Framework for Artificial Intelligence (March 20, 2026): White House legislative blueprint with preemption language
18. Executive Order 14365 (December 11, 2025): “Ensuring a National Policy Framework for Artificial Intelligence”
19. Cato Institute analysis: “Trump’s AI Framework: Federalism at Risk” (2026)
20. Brookings Institution: “The Empty National AI Policy Framework: Who Is in Charge?” (2026)
21. Bipartisan Policy Center: “State AI Laws and Federal Preemption: Finding Balance” (2026)

Federalism and Preemption Precedent

22. New York v. United States, 505 U.S. 144 (1992): Anti-commandeering doctrine — federal government cannot compel states to enforce federal law
23. Printz v. United States, 521 U.S. 898 (1997): Extended anti-commandeering to executive branch officials
24. Murphy v. NCAA, 584 U.S. 453 (2018): Anti-commandeering applies to affirmative prohibition on state legalization
25. Arizona v. United States, 567 U.S. 387 (2012): Field preemption analysis in immigration context
26. Gade v. National Solid Wastes Mgmt. Ass’n, 505 U.S. 88 (1992): Field preemption when federal regulation is “so pervasive”
27. Rice v. Santa Fe Elevator Corp., 331 U.S. 218 (1947): Presumption against preemption in traditional state authority
28. Cipollone v. Liggett Group, Inc., 505 U.S. 504 (1992): Conflict preemption framework
29. Bates v. United States, 522 U.S. 23 (1997): Express preemption analysis
30. Crosby v. National Foreign Trade Council, 530 U.S. 363 (2000): Conflict preemption where state law obstructs federal objectives

State AG and Legislative Response

31. Colorado Governor Jared Polis statement on AI preemption (March 2026)
32. California Attorney General guidance on AI vendor compliance (2025)
33. National Association of Attorneys General (NAAG) AI working group coordination (2025-2026)
34. Rep. Josh Gottheimer (D-NJ) comments on preemption concerns (March 2026)
35. Senator Maria Cantwell (D-WA) statements on federal AI standard requirements (March 2026)
36. House Democratic Caucus AI listening sessions announcement (March 2026)
37. TechNet position paper on federal AI preemption (2026)
38. US Chamber of Commerce AI governance recommendations (2025)
39. AI Now Institute: “State AI Laws: Progress and Preemption Threats” (2026)
40. Electronic Frontier Foundation (EFF): “Preemption Would Undermine State AI Protections” (2026)

—

Status: Draft | Citation count: 40 | Next: Part 3 — The Copyright Deferral