OpenAI предлагает «обратный федерализм»: законы штатов США как основа национальных правил ИИ

In July 2026, OpenAI published a policy article on artificial intelligence regulation in the United States on its corporate blog. The company proposes a "reverse federalism" approach: laws passed by individual states should become the building blocks of a nationwide framework for safe and democratic AI, rather than an obstacle to it.

What "reverse federalism" is

"Reverse federalism" is a bottom-up model of regulation: instead of the federal center handing down ready-made rules to the states, the states themselves are the first to pass AI laws, and a single national standard gradually emerges from their practice. In classic federalism the direction is the opposite: Washington sets the framework and the 50 states fit into it. OpenAI, as the publication suggests, proposes to recognize the states' legislative activity as a resource for building federal rules, not as fragmentation that needs to be stopped.

• The approach is called "reverse federalism" — a term from OpenAI's blog
• According to the National Conference of State Legislatures (NCSL), AI bills were considered in all 50 states in 2025
• In September 2025, California passed SB 53, a law on transparency of frontier AI models
• Colorado and Texas approved their own AI laws taking effect in 2026

The debate over the division of powers escalated in the summer of 2025, when the US Senate voted 99 to 1 to strip from the budget package an amendment imposing a ten-year moratorium on state AI regulation. The attempt to freeze state laws failed, and the question of who sets the rules for the industry remained open — and it is precisely this question that OpenAI now proposes to answer.

Why OpenAI is betting on the states

OpenAI views state laws as a proving ground where rules for future federal legislation are road-tested: successful approaches can be scaled to the whole country, while unsuccessful ones can be weeded out before they become nationwide. This is a notable shift in emphasis: as recently as 2025, major AI companies, including OpenAI, lobbied for the priority of federal rules and for protection from a "patchwork" of state requirements.

"State laws help build a national framework for safe and democratic AI,"

OpenAI's blog says.

The phrase "democratic AI" is no accident here: OpenAI regularly uses it in public documents, contrasting the development of technology in democratic countries with authoritarian models of control. The company presents regulation through elected state legislatures as part of that same line.

How states already regulate AI

Over the past two years, the states have become the main arena of AI regulation in the US. California's SB 53 law required major developers of frontier models to publish safety protocols and report incidents; Colorado and Texas approved their own AI acts, and dozens of other states adopted targeted rules — from deepfake labeling to rules on using AI in hiring.

It is precisely this body of laws that OpenAI proposes to treat as the blueprint for a national framework: the federal legislator only needs to synthesize the state rules that work, instead of writing rules from scratch. As long as Congress has no unified AI law, "reverse federalism" effectively describes a reality that has already taken shape — and gives it the status of a deliberate strategy.

What this means

OpenAI publicly acknowledges that a single federal AI law will not appear in the US anytime soon, and instead of fighting the patchwork of state laws it proposes to turn it into the foundation of national rules. For comparison: the European Union took the opposite path — a single AI Act applies at once to all countries of the union; the US, if OpenAI's forecast proves correct, will get a national framework assembled from state laws. For developers and businesses this is a signal: compliance will have to be built incrementally — from the requirements of California and Colorado to a future federal standard.