Across the United States—from state capitols to Washington and international forums—a coordinated approach to governing the most capable AI systems is taking shape. OpenAI has described the dynamic as “reverse federalism”: states adopting similar legal frameworks that converge toward a shared baseline and, in turn, help form a de facto national standard.
California, New York, and most recently Illinois have enacted frontier AI safety laws that embed democratic oversight into deployment. California established a core disclosure framework, New York demonstrated how that approach can transfer across jurisdictions, and Illinois added requirements for independent verification of key disclosures.
Why alignment matters
According to advocates, an aligned national standard is the best way to ensure safety while preserving U.S. innovation. A consistent approach is argued to:
- prevent a patchwork of conflicting state rules that could slow innovation and complicate enforcement; and
- position the United States to translate a national standard into a U.S.-led international framework grounded in democratic values.
The authors caution against performative measures and chaotic, divergent state regulation, which they say would undermine a durable safety regime and divert developer resources—especially at start-ups and small companies—away from safety work.
Core elements for state coordination
The article identifies a set of core elements around which states should align—elements that California, New York, and Illinois have begun to adopt together:
- clear public disclosures and accountability measures;
- independent audits or verification of certain disclosures; and
- security standards, incident reporting, and whistleblower protections.
Policymakers are also advised to avoid mission creep: states should not be expected to shoulder significant national security decisions or conduct extremely technical reviews that federal experts, with access to classified systems and greater resources, are better equipped to handle.
Federal role and testing framework
Work is also proceeding at the federal level. The Trump Administration is reported to be working with technical and national security experts on a framework for U.S. government testing of the most capable AI models—establishing testing standards, timelines, and processes—and OpenAI is engaged in discussions with the Administration, industry peers, business groups, and other stakeholders.
The article highlights the Center for AI Standards and Innovation (CAISI), created under President Biden and reinforced under President Trump, as a potential durable federal capacity to evaluate advanced models and shift frontier safety toward prevention rather than primarily after-the-fact accountability. The authors say federal legislation should clarify how CAISI interacts with other parts of government and its role in testing.
The Administration’s stated goal is to have this federal testing framework in place by early August. The current reality—models being tested before the federal framework is complete—has emphasized the need for consistent and repeatable approaches at both state and federal levels so that advanced models can reach government agencies, critical infrastructure defenders, allies, and other trusted partners promptly and safely.
Congressional and international developments
Congress is also active: lawmakers in both chambers and on both sides of the aisle, including Representatives Jay Obernolte and Lori Trahan, have proposed federal frameworks. While draft proposals are imperfect, the article views them as productive steps and believes many provisions merit support.
An international dimension is already emerging. At a recent G7-related discussion that included Brazil, Egypt, India, Kenya, and Korea, CEOs of leading frontier labs discussed the need for a global framework. Following that meeting, OpenAI CEO Sam Altman proposed in the Financial Times a "U.S.-led international forum" to establish accepted standards, provide impartial capability and risk analysis, and make technology available to adherent nations and companies. Google DeepMind CEO Demis Hassabis has also published related ideas.
The authors argue that bipartisan federal legislation would provide a strong basis for a U.S.-led international effort.
Conclusion
Momentum for a coordinated approach to frontier AI safety exists at multiple levels: state laws are converging, the executive branch and Congress are building toward a national framework, and global leaders are beginning to discuss international standards. If these efforts build on one another, the United States could lead the development of a global AI framework rooted in democratic values. The article concludes that this democratic alignment is the approach most likely to prioritize safety in the deployment of advanced AI systems.
