AI Security: The U.S. Moves Toward a Unified National Framework

The United States Unites for AI Security

In a concerted effort to enhance the security of artificial intelligence (AI), the United States is witnessing a collaboration between state and federal initiatives. This movement aims to establish a national standard that could potentially serve as a global model, under the leadership of the United States.

State Initiatives

Influential states such as California, New York, and Illinois have recently taken significant legislative actions to secure cutting-edge AI. These actions contribute to the creation of a common framework for governing the most advanced AI systems. This trend, often referred to by OpenAI as "reverse federalism," sees states playing a crucial role in establishing shared standards. By aligning with federal efforts, these state initiatives are paving the way for a national standard that could ultimately influence a global framework for AI led by the United States. The goal is to ensure that critical decisions regarding AI security are made by democratic governments rather than private entities.

Towards a National Framework

The idea of a national framework for AI is based on the principle that the United States must maintain its leadership in innovation. A national framework would avoid a patchwork of regulations that could hinder AI development. Such a standard would ensure that the United States remains at the forefront of innovation while protecting critical infrastructures from potential threats. Furthermore, it would facilitate the creation of a global approach for the secure deployment of AI, grounded in democratic values. To achieve this goal, close collaboration between states, the federal government, and international discussions is essential. A performative or symbolic approach will not suffice; it is crucial to avoid regulatory chaos at the state level, which could undermine a coherent security strategy.

Key Elements for State Alignment

For success, states must align on essential elements, as California, New York, and Illinois have done. These elements include a documented security framework with risk assessments for cutting-edge models, the reporting of serious security incidents, and governance through independent audits. These states have integrated democratic oversight into the deployment of advanced AI. California has established a central disclosure framework, New York has demonstrated the adaptability of this approach, and Illinois has mandated independent verification of key disclosures. While legislation may include additional provisions to secure the necessary votes, these elements are crucial for creating a de facto national standard through reverse federalism. Without this discipline, there is a risk of policy inflation and regulatory complexity that could divert valuable resources from developers, particularly startups and small businesses.

Concerns About National Security

Policymakers must also be cautious about expanding state missions. States should not be tasked with managing significant national security risks or making national security decisions on behalf of the entire country. These responsibilities require technical expertise and resources that only federal experts can provide, particularly through their access to classified systems and their ability to collaborate closely with national teams.

Federal Initiatives

At the federal level, the U.S. administration continues to collaborate with technical and national security experts to develop a testing framework for the most advanced AI models in terms of cybersecurity. This framework will define testing standards, timelines, and processes. OpenAI is actively participating in these discussions with the administration, partner companies, business groups, and other stakeholders to shape this effort. The goal is to have this framework in place by early August. A federal testing framework will enable advanced AI tools to be placed in the hands of the government, critical infrastructure defenders, allies, and other trusted partners, thereby strengthening democratic institutions and contributing to the construction of a democratic AI stack led by the United States.

Importance of Federal Legislation

Neither an indefinite federal process nor a patchwork of state laws will produce a coherent cutting-edge security regime. A national approach is necessary to ensure that the best testers evaluate the most advanced models and that trusted defenders have access to these tools quickly enough to stay ahead of malicious actors. Congress is also moving in this direction. Legislators from both chambers and both parties, including Representatives Jay Obernolte and Lori Trahan, have taken note of developments in the states and the executive branch and have proposed initiatives for a federal framework. While no proposal is perfect, these efforts are seen as productive, and many of their provisions are considered thoughtful and worthy of support. Senate and House leaders are also investing serious efforts into national governance and cutting-edge security proposals, and constructive conversations have taken place with them.

Developing a Security Framework

OpenAI's cutting-edge security plan outlines the essential elements of this framework. First, the federal government should lead the testing and evaluation of the most advanced systems. Cutting-edge AI raises national security and public safety questions that require technical expertise, resources, and access that states cannot fully replicate. This work should strengthen the Center for AI Standards and Innovation (CAISI), established under President Biden and bolstered under President Trump. CAISI can provide the sustainable federal capacity needed to assess advanced models and guide cutting-edge security 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 the government and what role it should play at the center of testing.

Second, companies developing the most advanced systems should meet clear requirements, including independent audits, incident reporting, strict security standards, and protections for whistleblowers. Third, federal and state efforts should mutually reinforce each other. State laws will not all be identical, and we look forward to working with policymakers across the country to ensure they enhance security while maximizing the economic benefits of AI. States should also continue to serve as laboratories of democracy in areas beyond cutting-edge security, including youth protection, energy and environmental policy, as well as education and AI literacy.

The Need for National Legislation

A federal framework remains essential. Cutting-edge AI raises national security, economic competitiveness, and public safety questions that ultimately require national standards, national capabilities, and national institutions to support democratic AI.

International Framework

National legislation is also crucial for a U.S.-led international framework for AI standards. This idea was discussed at the G7, with countries like Brazil, Egypt, India, Kenya, and South Korea, where CEOs of leading cutting-edge labs discussed the need for such a framework. Following this meeting, OpenAI CEO Sam Altman proposed in the Financial Times an "international forum led by the United States that establishes accepted standards, provides expert and impartial analysis of capabilities and risks, and makes technology available to nations and businesses that participate and comply with the rules." This week, Google DeepMind CEO Demis Hassabis also put forward thoughtful ideas in a new paper. Federal legislation—necessarily bipartisan—would provide a solid foundation for this international effort.

Momentum is now visible at all levels. States are establishing common approaches. Congress and the executive branch are building toward a national framework. And global leaders are beginning to discuss international standards. If everyone supports each other, the United States can lead the development of a global framework rooted in a democratic vision for AI. This democratic alignment approach for AI is one that truly prioritizes security.