Last updated: September 15, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely
Big AI sets out its terms for regulatory capture the same way any lobbied industry does: not by asking to be left alone, but by asking to write the rules itself. In 2026 that's playing out in public — record federal lobbying spend, a White House framework pushing Congress to preempt state AI law, and the two biggest labs publicly disagreeing on how far that preemption should go.
Short answer: "Regulatory capture" here means AI labs shaping the rules meant to govern them, rather than just opposing regulation outright. In H1 2026, Anthropic spent $3.53M and OpenAI $2.22M lobbying Congress, largely over whether federal law should preempt state AI rules. Check who's asking for what, and whether it's a floor or a ceiling, before trusting a headline.
In my testing of this story, I skipped the press releases and went straight to the primary documents: the White House's actual legislative recommendations PDF, the R Street coalition letter that preceded it, and the federal lobbying disclosures that show where Anthropic and OpenAI's money went. The picture that emerges is messier than "Big AI wants zero rules." Both labs lobbied hard, but they don't agree on what they're lobbying for, and at least one of them spent 2026 publicly supporting a state law most of the industry lobbied against. Here's how to read past the headline version.
What you need to know first
You don't need a law degree, but you do need three pieces of vocabulary. "Preemption" means a federal law overriding state laws on the same topic — Congress hasn't passed one for AI yet, only proposed it. "Floor vs. ceiling" describes whether a federal rule sets a minimum standard states can exceed (a floor) or a maximum states can't exceed (a ceiling) — this is the actual fight, not "regulation vs. no regulation." And "lobbying disclosure" refers to the quarterly LD-2 filings companies must submit under the Lobbying Disclosure Act, which is where the dollar figures below come from — not estimates, filed numbers.
Step-by-step: how to tell when Big AI is writing its own rules
1. Read the actual ask, not the press release
The White House released its National Policy Framework for Artificial Intelligence on March 20, 2026, built on the earlier Executive Order 14365. Its Section VII asks Congress to block states from regulating three specific things: AI development itself, the use of AI for activity that would already be lawful without AI, and developer liability for unlawful third-party conduct involving their models. That's narrower than "no state AI laws" — it's a targeted carve-out, and the difference matters for what your state can still do.
2. Follow the money, not the messaging
Anthropic spent $3.53 million lobbying the federal government in the first half of 2026, more than double what it spent in all of 2025 ($3.1 million), with $1.97 million of that in Q2 alone. OpenAI spent $2.22 million over the same six months, up sharply from the prior half. Combined Q2 spend across both companies hit $3.17 million, a 23% jump quarter over quarter. Anthropic's Q2 filings listed export controls, cybersecurity, and AI safety standards as focus areas; OpenAI listed copyright, cloud infrastructure, and privacy.
3. Check whether the ask is a floor or a ceiling
This is where the two labs split. A coalition letter organized by the R Street Institute on January 28, 2026 pushed for "time-limited preemption" of state AI rules while Congress writes something permanent — a ceiling. OpenAI has instead argued for what reporters have called "reverse federalism": build federal law on top of strong existing state laws rather than overriding them. Anthropic's public position is closer to a floor — it says it can accept preemption, but only if the federal replacement is at least as strong as the state law it replaces, not weaker.
4. Look for where "Big AI" isn't one thing
The clearest evidence this isn't a united front: Illinois. Gov. JB Pritzker signed SB 315, the AI Safety Measures Act, on July 6, 2026 — the first state law requiring independent third-party audits of frontier AI developers, plus mandatory disclosure of "catastrophic risk" (incidents likely to injure more than 50 people or cause over $1 million in damage). OpenAI and Anthropic both supported it, and it passed with only five "no" votes in the Illinois Senate. That's a strict state law two "Big AI" labs helped pass, the same year both were lobbying Congress about preemption elsewhere.
5. Map it to your own compliance timeline
None of the federal preemption proposals are law yet — they're recommendations Congress hasn't acted on, including rejecting preemption language in both the One Big Beautiful Bill Act and the 2026 NDAA. Illinois's SB 315 takes effect January 1, 2027, and is real regardless of what Congress does. If you operate in a state with an AI law already on the books, plan around that law, not around a federal proposal that may never pass.
Where OpenAI and Anthropic actually disagree
| OpenAI | Anthropic | White House framework (Mar. 20, 2026) | |
|---|---|---|---|
| Preemption stance | "Reverse federalism" — build on strong state laws | Preemption only as a floor, not a weaker ceiling | Preempt state rules on AI development, lawful AI-assisted activity, and third-party liability |
| H1 2026 federal lobbying | $2.22 million | $3.53 million | Not a lobbying spend — a policy document |
| Illinois SB 315 (audits, catastrophic-risk disclosure) | Supported | Supported | Not addressed |
| What it protects against | A 50-state patchwork of conflicting compliance rules | Preemption being used to erase stronger existing state protections | A "light-touch," uniform national standard |
If you only take one row from this table, take the second. Nearly $5.75 million in combined H1 2026 lobbying spend, filed under the Lobbying Disclosure Act, is a better indicator of where a company's actual priorities sit than any blog post about "responsible AI policy."
Example prompts you can copy
Use these to run your own version of this check on a bill or company statement:
- "Here's a company's public statement on [bill name]. Based on this text alone, is it asking for a floor (a minimum states can exceed) or a ceiling (a maximum states can't exceed)? Quote the specific language that tells you."
- "I operate in [state]. Summarize what AI-specific laws are already in effect there versus what's just been proposed at the federal level, and flag which one I should actually plan around."
- "Compare these two companies' public positions on AI preemption [paste both]. List only where they substantively disagree, not where they're using different words for the same thing."
- "Look up the Lobbying Disclosure Act filing for [company] for [quarter]. What specific issues did they list, separate from what their press office said they lobbied for?"
- "Given this state law is set to take effect on [date] and this federal proposal hasn't passed Congress, which one should shape my compliance plan for the next 12 months?"
Common mistakes to avoid
The biggest one: treating "Big AI" as a single actor with one lobbying position. OpenAI and Anthropic disagree publicly on preemption, and conflating them hides the actual fight. Second, assuming a federal framework is law the moment it's announced — the White House's March 2026 document is a set of recommendations to Congress, and Congress has already declined similar preemption language twice this year. Third, reading "preemption" as "deregulation" — the R Street letter and the White House framework both frame it as replacing a patchwork with a standard, not eliminating rules, even though the practical effect on a given state's protections can be the same. Fourth, ignoring state law because a federal proposal is in the news — Illinois's SB 315 takes effect January 1, 2027, with or without federal action. Fifth, trusting a company's stated position over its actual disclosed lobbying spend, which is a filed legal document, not marketing copy.
Tools that make this easier
You don't need a policy team to track this yourself. If you want the broader context on how AI companies' own usage policies interact with government rules and your workplace's, my guide to who actually sets the rules for AI breaks down all three layers. For the EU side of this same fight, what’s actually enforceable under the EU AI Act right now covers the parallel patchwork-versus-standard debate happening in Brussels. If you're building an internal policy so this doesn't matter as much day to day, start with how to write an AI policy for your business, and if disclosure and messaging rules are the part you're worried about, how AI regulation and messaging rules apply in practice goes deeper than this piece has room for. For general AI policy news as it breaks, I keep our AI news roundup updated weekly. And if the underlying question is really "which of these companies' tools should I even be using," Claude vs. ChatGPT and how to use ChatGPT are the practical starting points.
My take
"Regulatory capture" usually gets used as a slur, implying an industry bought its way out of rules entirely. What I found doing this research is narrower and more interesting: both labs want rules, just rules shaped around what they can already comply with, filed through the same legal lobbying process every regulated industry uses. Anthropic and OpenAI supporting Illinois's audit law while lobbying Congress about preemption elsewhere isn't a contradiction — it's a company deciding which fights are worth having in which venue. That's not a scandal by itself. The mistake is assuming their interests and yours line up by default. They usually don't, which is exactly why checking the actual filing beats trusting the press release.
Frequently Asked Questions
Big AI sets out its terms for regulatory capture: is this actually happening right now?
Yes, in the sense that both leading labs are actively lobbying to shape AI law rather than just opposing it. Anthropic and OpenAI together spent $5.75 million lobbying the federal government in H1 2026, largely over how far federal preemption of state AI law should reach.
Has Congress already preempted state AI laws?
No. The White House's National Policy Framework, released March 20, 2026, is a set of legislative recommendations, not a passed law. Congress has twice declined to include AI preemption language this year, in the One Big Beautiful Bill Act and the 2026 NDAA.
How long does it take to check whether an AI rule actually serves the industry that lobbied for it?
Checking one specific bill or statement against the company's actual lobbying disclosure usually takes 15 to 30 minutes once you know where to look — the filed LD-2 disclosures, not the company's blog.
What is the easiest way to tell if a company genuinely supports a rule or just tolerates it?
Check whether their public statement describes the rule as a floor (a minimum other rules can exceed) or a ceiling (a cap nothing can exceed). Genuine support for accountability tends to show up as a floor position; preemption asks are usually a ceiling.
Do OpenAI and Anthropic agree on AI regulation?
Not fully. Both supported Illinois's strict SB 315 audit law, but they diverge on federal preemption — OpenAI favors building on strong existing state laws, while Anthropic says it can accept preemption only if the federal replacement is at least as strong as what it replaces.