The US is advancing AI safety through state and federal action

The US is advancing AI safety through state and federal action

Through reverse federalism, states are aligning on AI safeguards as the federal government builds toward a national standard—laying the groundwork for a US-led global framework.

By Chris Lehane, Chief Global Affairs Officer

From state capitals across the country to Washington to international convenings, serious approaches to frontier AI governance are taking shape. Together, they are advancing a democratic vision for AI.

California, New York, and most recently Illinois have advanced frontier safety legislation that helps move the country toward a common baseline for governing the most powerful AI systems.

These efforts reflect momentum behind what OpenAI calls reverse federalism: states helping establish a shared direction through a common framework. As state-led work converges with efforts underway at the federal level, a national standard is beginning to take shape, laying the foundation for a US-led global AI framework. We believe the best way to ensure AI benefits the many, not just the few, is for critical decisions—starting with frontier safety—to be made by democratic governments, not solely by frontier labs.

Ultimately, the United States would be best served by a national framework grounded in that principle. But in the absence of one, states can move us there by passing laws that mirror one another. Step by step, they can create a de facto national standard.

And if safety is the priority, such a national standard is the best policy approach to ensure (a) the US continues to lead in innovation and not be undermined by a patchwork of regulations that would slow the ability to build out a global democratic AI stack that will optimize for the good guys having the tools they need to defend against bad actors seeking to misuse AI in areas like cybersecurity; and (b) put the US in the strongest position to take the national standard and create a global approach for the safe deployment of AI that is built on democratic values. If we want to create a safety framework that will truly prioritize getting in place a national and international standard—which is what will best support the safe deployment of AI—it will take hard work, a seriousness of purpose and a strategy that connects the states, the federal government, and the international conversation. A performative or virtue signaling approach will not get the job done here—chaos at the state level is not in the best interest of a durable approach to safety, it will only lead to more chaos.

For that approach to succeed, states should continue aligning on the core elements, as California, New York, and Illinois have. Broadly speaking, those elements are:

• A documented safety framework with risk assessments for frontier models and public disclosure of those assessments and their results.
• Reporting of serious safety incidents.
• Governance and accountability through independent, objective audits.

Taken together, these three states have built democratic oversight into frontier AI deployment. California established the core disclosure framework. New York showed the approach could be adopted across jurisdictions. Illinois complemented it by requiring independent verification of key disclosures.

We recognize that legislation often includes additional provisions to secure the votes needed for passage. But we believe these are the essential elements needed to create a de facto national standard through reverse federalism. Without that discipline, we risk policy creep and a patchwork of state laws that are difficult for regulators to enforce, confusing for consumers to navigate, and divert developer resources—especially at start-ups and small companies—that would be better invested in safety. In particular, and as the last few weeks have shown, as AI models become more and more capable, we need a coherent system that will allow us to get the tools into the hands of government, critical infrastructure, allies, and trusted partners.

Policymakers should also guard against mission creep. States should not be asked to manage significant national security risks (or be in effect making national security decisions on behalf of the whole country) or conduct highly technical reviews that are better handled by federal experts with the resources, expertise, access to classified systems, and capacity to work closely with our teams.

At the federal level, the Trump Administration continues to work with technical and national security experts on a framework for US government testing of the most capable AI models on cyber. That framework will establish testing standards, timelines, and processes. OpenAI is engaged in constructive discussions with the Administration, peer companies, business groups, and other stakeholders helping shape the effort.

The ongoing work on cyber evaluations illustrates why consistency matters. Today, and understandably, models are being tested before the federal framework is complete. As labs have worked through that process, one lesson has become clear: we need a consistent and repeatable approach at both the state and federal levels. That is essential if we want the most capable models in the hands of government, critical infrastructure defenders, allies, and other trusted partners. We appreciate the Administration’s goal of having this framework in place by early August.

A federal testing framework will help get advanced AI tools into the hands of government, critical infrastructure defenders, allies, and other trusted partners. In doing so, it will strengthen democratic institutions and help build a US-led democratic AI stack. Now is the time to use America’s innovation lead in support of democratic AI.

Neither an undefined federal process nor a patchwork of state laws will produce a coherent frontier safety regime. We need an approach that ensures the best testers evaluate the most capable models—and that trusted defenders gain access to those tools quickly enough to stay ahead of malicious actors.

Congress is moving as well. Lawmakers in both chambers and on both sides of the aisle—including most recently Reps. Jay Obernolte and Lori Trahan—have taken note of developments in the states and the executive branch and have put forward proposals for a federal framework. No discussion draft with a realistic path to passage will be perfect. But we view this work as a productive step forward and believe many of its provisions are thoughtful and worthy of support.

We are likewise encouraged that Senate and House leaders are investing serious effort in national governance and frontier safety proposals, and we’ve had constructive conversations with many of them. Given the size and influence of the states that have already enacted directionally aligned laws, incorporating those approaches into federal legislation should make it easier to establish a single national frontier safety regime.

OpenAI’s frontier safety blueprint lays out what we believe are the essential elements of that framework.

First, the federal government should lead the testing and evaluation of the most advanced systems. Frontier AI raises national security and public safety questions that require technical expertise, resources, and access that no state can fully replicate.

That work should strengthen the Center for AI Standards and Innovation (CAISI), created under President Biden and strengthened under President Trump. CAISI can provide the durable federal capacity needed to evaluate advanced models and shift frontier safety toward preventing harm before it occurs, rather than relying primarily on accountability afterward. Any federal legislation should carefully consider how CAISI should work with the rest of government and what role it should play at the center of testing.

Second, companies developing the most capable systems should meet clear requirements, including independent audits, incident reporting, strong security standards, and