The Patchwork Problem: New York, California, and the Case for Cohesive AI Rules
Two of the largest technology markets in the United States are writing their own rules for artificial intelligence, and they are not writing the same rules. California's frontier AI transparency law has been in force since January 1, 2026, and Sacramento sent roughly thirty more AI bills to the Governor this session. New York's amended RAISE Act takes effect January 1, 2027, with its own reporting clocks, penalties, and requirements. For the founders and AI companies we represent, the question is not whether AI will be regulated. It is whether a company building one product for one national market will face coherent, predictable rules or a growing patchwork of conflicting ones. Innovation depends on the former, whatever form it ultimately takes.
Timing matters. In a September 9 essay, OpenAI's Chris Lehane argued that "the AI policy window is open" and that "the greater risk now is waiting too long" to act. Whether or not one shares that company's policy preferences, the point applies to every AI business: the rules written in the short terms likely to shape the industry for much longer.
California: from transparency to verification
California's Transparency in Frontier Artificial Intelligence Act (SB 53) requires large frontier developers to publish safety frameworks and report critical safety incidents, with penalties of up to $1 million per violation. The state is now building on that foundation. In September 2026 the Governor signed SB 813, creating a state process for independent verification organizations, and AB 1405, creating a registry of approved AI auditors. He also signed measures on companion chatbots (SB 1119, SB 867) and AI-generated performers in advertising (SB 1050).
The direction of travel is clear. A September 18 executive order accelerates those oversight laws and asks experts to consider on-site independent auditors at frontier labs and an "AI kill switch" for frontier models. Several other bills, including rules on automated employment decisions (SB 947) and customer-service chatbots (AB 1609), await action before the Governor's September 30 deadline.
New York: similar in spirit, different in detail
New York Governor Hochul signed the RAISE Act in December 2025 and signed amendments on March 27, 2026 that move it closer to California's model. The law covers developers with more than $500 million in revenue that train models above 10°26 integer or floating-point operations. It takes effect January 1, 2027.
Alignment is welcome, but the differences matter to anyone who must comply with both statutes:
Requirement | California (SB 53) | New York (RAISE Act) |
Critical incident reporting | 15 days | 72 hours; 24 hours if imminent risk of harm |
Maximum penalty | $1 million per violation | $1 million first; $3 million subsequent |
Regulator rulemaking | None expressly granted | New DFS office with broad rulemaking power |
Ownership disclosure | Not required | Beneficial owners filed with DFS every two years |
The New York Department of Financial Services can also add reporting and publication requirements by rule. A company in full compliance today may face new obligations tomorrow, set by an agency rather than a legislature.
Washington: the federal debate
The federal government has signaled that it prefers a single framework. Executive Order 14,365 (December 11, 2025) created a Justice Department AI Litigation Task Force to challenge state AI laws. On March 20, 2026, the White House asked Congress for a "minimally burdensome national standard" that would preempt unduly burdensome state rules while leaving states their traditional authority over child safety, fraud, and consumer protection.
In Congress, a bipartisan House group led by Representatives Obernolte and Trahan released the Great American Artificial Intelligence Act as a discussion draft on June 4, 2026. It would impose national safety-framework and audit duties on large frontier developers and preempt state laws regulating model development for three years. The draft had not been formally introduced as of late September, and more than 200 state lawmakers have opposed its preemption clause.
Whether it is Federal legisation, nationwide coordiantion, or system of standards, what matters is the outcome: rules that are cohesive enough that a developer can build once, comply once, and keep moving forward. Frontier models are trained once and deployed nationally, and divergent incident-reporting clocks, penalty schedules, and regulators add cost without adding safety. Those costs fall hardest on startups, which lack the compliance budgets of the largest companies. A national framework is one route to cohesion; harmonized state laws, of the kind New York's amendments began to pursue, are another.
Industry itself has begun to sketch what cohesion could look like. OpenAI's essay describes a "reverse federalism," in which state laws such as California's SB 813 and AB 1405 set working baselines that Congress can later codify nationally. It also calls for obligations that are capability-based and proportionate to a developer's resources and risk, and for compatible international approaches, since models and expertise cross borders. Several of these ideas travel well regardless of where one lands on preemption:
Proportionality. Obligations should scale with capability and risk, so that early-stage companies are not saddled with rules designed for the largest labs.
Competition. Safety rules should not entrench incumbents or push development overseas.
Interoperability. State, federal, and international requirements should be compatible enough that one compliance program can satisfy them all.
Safety as an enabler. In Lehane's words, safety "is what allows progress to go further and benefit more people." Clear rules build the public trust that adoption depends on.
What AI companies should do now
Until Congress acts, the state laws are the law, and companies should plan to comply with them. We recommend three steps:
Map potential exposure. Confirm whether your revenue and training compute bring you within SB 53 or the RAISE Act, and whether your products touch California's new chatbot, advertising, or employment rules.
Build to the stricter standard. One safety framework and one incident-response process designed for New York's 72-hour clock can usually satisfy both states.
Engage in the policy debate. The federal draft remains open for comment, and New York's DFS will write rules. Companies that participate now will help shape the standards they later must meet.
