President Trump is : if the U.S. wants to lead on AI, it can’t have 50 disparate rules regulating the technology. As he and his administration continue to for AI preemption, 91快活林 joins their call for a clear, national standard to address concerns about AI risks – ensuring industry can continue to innovate and businesses and consumers have trust in new technology. The right policies will pave the way for all Americans to benefit from the job and economic growth, energy efficiencies, improved health care outcomes, and better government service delivery that AI will bring.
At 91快活林, we represent the full AI stack of companies that design, build, and deliver the models, hardware, software, data centers, cloud services, networking solutions, and applications. Our members’ AI investments and products are creating tremendous economic benefit in the U.S. and in markets around the world.
But the specter of conflicting state level AI requirements is quickly becoming a reality –threatening companies’ ability to advance this transformative technology. In 2025, over 100 AI bills became law in 38 different state capitols. This patchwork of competing AI regulation creates confusion for consumers and burdensome compliance costs for businesses of all sizes.
As the Trump Administration and Congress work towards providing the necessary federal clarity, here are examples of the vast overreach and divergence of state level AI mandates:
Broad and inconsistent definitions. State laws that are proposing both problematic and conflicting definitions will make it difficult for businesses to comply with every state’s rule and consumers to understand what AI systems are subject to the law.
- Artificial Intelligence System. would adopt the nearly verbatim (“a machine-based system that infers from inputs to generate outputs… that influence physical or virtual environments”). By contrast, ,鈥�enacted this year, adds additional requirements that AI systems must “abstract perceptions into models” and use “automated inference” to formulate actions.
- Frontier/Foundation Models Compute Thresholds. California’s recently enacted and New York’s pending both use compute thresholds exceeding鈥�10^26 integer operations, but differ on additional criteria; California uses a鈥�$500M annual revenue threshold, while New York uses a鈥�$100M compute cost threshold. Compute is not necessarily indicative of risk, and a predetermined compute threshold is also not future proof, given the way in which technology evolves. Additional differing revenue thresholds across states adds to the confusion.
- Chatbots.鈥�would define an “AI chatbot” broadly as any software that simulates human conversation, while New York’s enacted S3008C provides a completely different definition covering specific capabilities. Differing definitions of AI chatbot risks over-inclusion, undermining educational potential and efficiency gains through these tools.
State disclosure mandates are operationally burdensome and all over the map. States have rapidly expanded鈥痙isclosure requirements with varying degrees of disclosure creating operational burdens. These approaches also vary considerably across jurisdictions. The information consumers receive about AI use should not depend on their zip code and conflicting disclosure requirements could harm the ability to offer some AI services nationwide.
• Overly Broad Requirements. 鈥�would require voluminous disclosures, including the categories of data used in automated profiling, the decision logic employed, any evaluations for fairness or bias, and clear instructions for consumers to opt out of automated processing.
Highly Prescriptive and Operationally Burdensome. regulations impose some of the most rigid and expansive disclosure obligations in the country. Businesses must satisfy an extensive list of formatting and delivery mandates which in practice require constant redesign. These requirements create a prescriptive compliance framework that is difficult to operationalize and adds substantial friction to deploying and updating AI systems in California.
Fragmented responsibilities across the AI value chain.鈥�Developers, deployers, and integrators could face different obligations depending on the state, potentially creating a chilling effect on innovation.
- Developer vs. Deployer Obligations Diverge Sharply. Some states now have laws on the books that place heavier burdens on developers, such as the and California’s SB 53. Other states have proposed bills, such as or , that focus primarily on deployer-facing duties. Pennsylvania’s would impose sweeping liability on AI deployers (including criminal liability in some cases) without adequately distinguishing between developers, deployers, and integrators, holding all actors to the same standard across a wide range of potential harms.
- Lack of Consistency in Requirements.鈥�Some states’ bills would require assessments (e.g., ), while others like New York’s RAISE Act emphasize labeling and notification.
- Integrator Obligations Are Emerging but Undefined.鈥�A few proposals (e.g., Texas HB 149) include integrator duties, but definitions and expectations remain unclear.
Fragmented definitions, divergent risk classifications, inconsistent disclosure mandates, and misaligned responsibilities across the AI value chain create unnecessary barriers to AI innovation, without providing the public with the consistency and certainty they need to reap the benefits of AI adoption. We urge President Trump and Congress to prioritize U.S. leadership and establish a clear, preemptive national framework that ensures that businesses of every size can build and deploy AI responsibly without navigating fifty different regulatory regimes.