OpenAI wants one rulebook before states write fifty of them

OpenAI published a public policy agenda for AI covering frontier safety, youth protection, education, workforce transition and infrastructure. The real story is an attempt to keep AI rules legible before fragmented regulation turns deployment into paperwork archaeology.

OpenAI frames frontier safety as a federal issue, not a voluntary promise

OpenAI says safety for the most advanced general-purpose models is a national security and public safety issue. It names cyber, CBRN and public safety risks and supports aligned state frameworks including California SB 53, the New York RAISE Act and Illinois SB 315.

The company also calls for a US federal framework. It wants the Center for AI Standards and Innovation strengthened as the federal government's primary institution for frontier AI safety, with evaluations of the most capable models and an independent assessment ecosystem. After that, OpenAI supports preemption of state laws covering the same risks.

The agenda also covers teen safety, age assurance, parental controls, audits, synthetic CSAM, deepfakes, provenance signals such as C2PA, AI literacy in schools, workforce training and datacenter policy. On infrastructure, OpenAI says datacenters should pay their incremental electricity costs and that Stargate sites should have community plans.

For enterprise buyers, the useful part is less regulatory fragmentation

The practical point for enterprise teams is not that OpenAI has a position on everything. It is that a major AI vendor is translating governance into categories that legal, security and procurement teams can process: evaluations, incident reporting, audits, provenance, parental controls, workforce transparency and datacenter cost allocation.

That matters outside the US too, even though the agenda is mainly written for US policy. It shows how large AI suppliers will frame compliance for governments, schools and regulated companies. If these categories stick, they will appear in security questionnaires and contracts.

The second layer is sharper. OpenAI does not only want rules. It wants rules at the right layer. A federal framework is more predictable for the company than a mosaic of state laws, local bans and incompatible audits.

Public interest and lobbying share the same page

Many proposals are sensible, but readers should not confuse a policy agenda with an independent standard. OpenAI supports transparency, audits and safety evaluations, while also explicitly preferring harmonization and federal preemption. That reduces the number of places where regulators can push back.

Availability also needs care. This agenda does not mean specific protections or product features are available in the EU or Czechia. It is a list of policy priorities and commitments, not a product changelog.

Audits will decide more than polished principles

Three signals are worth watching: how concrete US federal proposals become, whether CAISI gets evaluation capacity and what independent teen safety and frontier safety audits look like. Without those details, the agenda remains a well-written wish list.

The second test is infrastructure. If OpenAI says datacenters should not raise electricity prices for nearby households, the important question is whether large loads actually pay their incremental costs through tariffs.

Lilith's verdict

OpenAI is not writing a manifesto for safer AI. It is fighting for access to the legislative process before the forms that give schools, agencies and datacenters their stamp of approval get locked without it.