Should States or the Federal Government Set AI Safety Rules?

Hi everyone, it's Shiichan! Today's story is a bit meatier, but important: who actually gets to write the rules for AI safety.

OpenAI News openai.com

What was announced?

OpenAI News published a piece by Chris Lehane, OpenAI's Chief Global Affairs Officer, about an idea called reverse federalism. In areas where there's no national rule yet, several states passing similar laws can end up building a de facto national standard on their own.

According to the piece, three states have already enacted frontier AI safety legislation: California, New York, and Illinois. California set up the foundational disclosure framework, New York showed the approach could transfer to another state, and Illinois added independent verification requirements for key disclosures.

These laws share three common elements:

• A documented safety framework with risk assessments and public disclosure
• A mechanism for reporting serious incidents
• Independent audits that ensure governance and accountability

Lehane notes that while each law may include extra provisions needed to get passed, these three form what he calls 'the essential elements needed to create a de facto national standard.'

Why it matters

When AI safety rules differ wildly from state to state, it's confusing for companies and users alike, and it's easier for gaps to slip through. Lehane's argument is that instead of waiting for a perfect federal law to appear, stacking up similar state laws can get you to a real national standard faster.

The piece also touches on federal-level movement: the Trump Administration is developing a testing framework for advanced AI systems focused on cybersecurity risks. OpenAI supports building consistent evaluation standards and timelines here, arguing that consistency is what makes it possible to get capable models into the hands of government agencies, critical infrastructure defenders, and allies.

What changes

The piece praises congressional efforts too, including proposals from Representatives Jay Obernolte and Lori Trahan, while noting that 'no discussion draft with a realistic path to passage will be perfect.' If state laws and a federal framework line up well, AI companies won't have to juggle a patchwork of state-by-state standards, and safety expectations become more visible to users as well.

The story doesn't stop at the US border, either. The piece references recent G7 discussions and Sam Altman's proposal for a US-led international forum that would establish accepted standards and make capable technology available to countries and companies that meet them. The piece's overall stance is that democratic governance is central to deploying AI safely.

Wrap-up

• OpenAI's Chris Lehane introduces the idea of 'reverse federalism' for AI safety rules
• California, New York, and Illinois have already enacted frontier AI safety laws
• The shared core: disclosed safety frameworks, serious incident reporting, and independent audits
• Federally, a cybersecurity-focused testing framework is in development, alongside bipartisan congressional efforts
• The piece also references G7 discussions and a proposed international forum tied to shared standards

Worth a read if you follow AI policy and governance, or if you're wondering how a company should think about navigating AI rules that differ state by state.