States and the federal government are pushing in the same direction: to create clear rules for the most powerful AI systems. Why does it matter? Because without a common standard, we risk a legal jigsaw that confuses companies, weakens security, and delays defenders and allies from accessing critical tools.
What's happening and why it matters
In recent months, places like California, New York, and Illinois have passed laws that set a minimum floor for oversight and safety of what's called “frontier AI,” the most advanced models. OpenAI describes this as “reverse federalism”: states set a path that, if aligned, can become a de facto standard before a unified federal law exists.
Sounds complicated? Think of it like synchronizing traffic lights in a city: if every district does what it wants, you get jams. If they move together, traffic flows and everyone benefits.
Key elements that states are aligning on
These are the essential components that are converging across state laws and that, according to the piece, should form the basis of a national standard:
- A documented safety framework with risk assessments for frontier models and the public release of those results.
- Mandatory reporting of serious safety incidents.
- Governance and accountability through independent, objective audits.
California laid the groundwork for disclosure requirements. New York showed that the approach can be adopted across multiple jurisdictions. Illinois added independent verification of disclosures. Together, these steps introduce democratic oversight over the deployment of very capable systems.
Without discipline and convergence, clear warning: a mosaic of state laws can be worse than having no rules.
Role of the federal government and the coordination needed
OpenAI and other actors ask that the federal government take technical leadership, especially on testing and evaluating the most capable systems. Why? Because there are national security issues and access to classified resources that states simply can't handle.
The proposal includes strengthening the Center for AI Standards and Innovation (CAISI) to coordinate testing, prevent harms before they happen, and standardize processes. It also calls for clear requirements for companies: independent audits, incident reporting, high safety standards, and protections for whistleblowers.
The administration and Congress are moving in parallel. There are bipartisan talks and draft bills that already incorporate ideas that came up in the states. The goal: to avoid states doing technical evaluations that should be a federal responsibility.
What this means for innovation and the world
If a coherent federal standard is achieved that reflects what several states already do, the U.S. could maintain its leadership in innovation while building a democratic framework that serves as an international reference.
In global forums like the G7 and in meetings with countries such as Brazil, India, or South Korea, discussions are underway on how to turn that national standard into rules accepted worldwide. The idea is simple: a democratic, transparent framework makes collaboration with allies easier and reduces the space for malicious use.
Risks and precautions
There are two clear warnings in the piece:
- Don’t ask states to take on responsibilities that require federal capabilities, like national security decisions or deep technical reviews.
- Avoid performative initiatives that aim for political applause instead of effective measures.
In practice, this means coordinating tests, preventing regulatory fragmentation, and prioritizing fast, secure access for those who need tools to defend critical infrastructure and governments.
The transition described isn’t automatic. It will take work, dialogue between states and the federation, and political will. But the window is there: states acting as laboratories, a Congress taking note, and a federal government with the technical capacity to institutionalize a standard.
The decisions made now will define not only domestic safety, but the U.S. ability to lead an international approach to AI aligned with democratic values.
Source original
https://openai.com/index/advancing-ai-safety-through-state-and-federal-action