OpenAI's Newsom Letter: The AI Governance Fight Companies Already Lost

OpenAI Asked California to Step Aside. Six Weeks Later, California Signed a Frontier-Model Law.

On August 12, 2025, OpenAI published an open letter to Governor Gavin Newsom urging California to harmonize its AI rules with emerging national and global standards. The company pointed to roughly 1,000 AI-related bills moving through state legislatures that year and warned of a fragmented compliance burden. The ask was straightforward: let federal frameworks lead, and treat compliance with them as sufficient in California.

But here is the part the coverage missed. Six weeks later, on September 29, 2025, Newsom signed SB 53, the Transparency in Frontier Artificial Intelligence Act, into law. The bill imposes state-specific safety-framework and disclosure obligations on large frontier developers with more than $500 million in annual revenue. OpenAI did not get harmonization. It got a second California AI statute in twelve months.

This was not a lobbying failure in the ordinary sense. It was a signal that the harmonization theory of AI governance, the idea that a single light-touch federal standard will displace state law, is not the direction of travel. Here is what happened, why it matters, and what companies building or deploying AI should do about it.

The Federal Preemption Bet Companies Made — and Why It Is Not Paying Off

OpenAI's harmonization argument rests on a specific policy bet: that Washington will produce one authoritative AI standard and that states will defer to it. That bet has a documented track record in 2025, and the record cuts against it.

The moratorium died 99 to 1

On July 1, 2025, the U.S. Senate voted 99-1 to strip a proposed 10-year moratorium on state AI regulation from the One Big Beautiful Bill Act. The House passed the amended bill on July 3, and it became law on July 4 without the preemption provision. A near-unanimous Senate declined to freeze state AI lawmaking. That is not a close call. It is a rout.

The federal standard is voluntary, not preemptive

OpenAI's letter cited its commitment to work with the Center for AI Standards and Innovation (CAISI), the body established within NIST in June 2025 as the successor to the U.S. AI Safety Institute. CAISI develops standards. It does not preempt state statutes. A voluntary standards body and a binding state disclosure law are different instruments with different legal force.

The distinction that matters: federal standards can guide, but only Congress can preempt, and Congress just declined to. Until that changes, state law is the operative law. Companies that build compliance programs around a hoped-for federal ceiling are building on a foundation the Senate already removed.

Why California Is the Standard, Not the Exception

AI governance in the United States is now a state-led project, and California is the anchor. Newsom vetoed the earlier and broader SB 1047 on September 29, 2024, criticizing it for applying stringent standards based on model size rather than deployment risk. One year later to the day, he signed SB 53, a narrower bill authored by Sen. Scott Wiener that survived by targeting the largest developers and focusing on transparency.

That sequence tells you how state AI law will develop. Not through a single sweeping statute, but through iterative bills that get recalibrated until they pass.

• SB 53 applies to large frontier developers. The $500 million revenue threshold means the biggest labs carry the compliance load first, but thresholds move over time.
• The obligations are disclosure-forward. Publishing an AI safety framework is now a legal requirement in California for covered developers, not a best practice.
• The mechanism is state civil enforcement. This is not a federal agency guidance document. It is enforceable state law with penalties.

For any company treating California as one jurisdiction among fifty, that framing is backwards. When roughly 1,000 AI bills are in motion nationwide, California's enacted text becomes the template other legislatures borrow. The practical reality is that complying with California is closer to complying with the emerging national baseline than complying with any voluntary federal framework.

What AI Companies Should Do Now

The harmonization debate will continue in Washington. Your compliance obligations will not wait for it to resolve. Here is where to focus.

Build to the strictest enacted state law, not the hoped-for federal floor

First, map your exposure against SB 53, not against CAISI guidance. If you are a frontier developer at or approaching the $500 million revenue threshold, the transparency and safety-framework obligations are live legal duties. Voluntary federal standards do not discharge them.

Second, treat published AI safety frameworks as regulatory documents. SB 53 requires covered developers to publish safety frameworks. A published framework is a representation. If your practices do not match your published document, you have created enforcement exposure. Draft it the way you would draft a securities disclosure.

Third, monitor the state bill pipeline as a leading indicator. With approximately 1,000 AI-related bills in state legislatures, the next binding obligation is more likely to originate in Sacramento, Albany, or Austin than in a federal rulemaking.

• Inventory every jurisdiction where you train, deploy, or offer AI systems.
• Identify which enacted or pending bills reach your revenue tier and model class.
• Align internal governance documentation with the strictest applicable standard.

Fourth, do not assume preemption is coming. After a 99-1 Senate vote against a moratorium, betting your compliance posture on federal displacement of state law is not a strategy. It is a hope.

Key Takeaways

• The harmonization ask did not land. OpenAI's August 12, 2025 letter urged California to defer to federal standards, and six weeks later Newsom signed SB 53, a state frontier-model law.
• Federal preemption is off the table for now. The Senate voted 99-1 on July 1, 2025 to strip the 10-year state AI moratorium from the One Big Beautiful Bill Act, and the final law contained no preemption.
• Voluntary federal standards do not discharge state duties. CAISI, established within NIST in June 2025, develops standards but does not preempt binding state statutes like SB 53.
• California is the operative baseline. SB 53 applies to large frontier developers with more than $500 million in annual revenue and requires published AI safety frameworks under enforceable state law.
• Build to the strictest enacted law. With roughly 1,000 AI bills in motion across state legislatures, compliance programs anchored to a hoped-for federal floor are exposed.