Investigative Analysis · AI Policy · Lobbying · Congress

11 Bills, $145 Million, and a Six-Week Recess: Inside the AI Lobbying War Nobody Voted On

OpenAI's president personally gave $25 million to kill state AI laws. Anthropic gave $20 million to save them. Both companies are publicly saying the opposite of what their money is doing. Meanwhile, a viral resignation post citing a "greater than 10% chance of extinction" is being used to rush a safety bill through Congress before the House leaves for six weeks. Here is every bill, who benefits, and what it's actually worth.

The safety billThe preemption warThe chip line-jumpThe anti-moat outlierThe extinction postThe six-week clock

September 23, 2026•19 min read•Subjects: OpenAI · Anthropic · Congress · FEC filings · 11 bills, 119th Congress

Primary sources: FEC filings · GovTrack · IAPP · NPR · CNBC · The NationMethod: Bill-by-bill breakdown + follow-the-money analysis

The Thesis

Eleven AI bills are moving through the 119th Congress right now. Only one has real momentum before the House leaves for a six-week election recess. The rest are a mix of long shots, quiet wins, and pure paperwork. But the pattern across all eleven is the same: whoever already has the most lawyers, the most compute, or the most lobbying budget tends to come out ahead — with two deliberate exceptions built specifically to work the other way.

This report grades each bill on one question a regular person never gets asked: does this law make it harder for a new company to compete, even if that was never the stated goal? That's a "regulatory moat." Five bills build one. Two are designed to tear one down. The rest barely move the needle — which itself is informative about who is and isn't spending money to shape them.

Five Terms Before You Read Any Further

Regulatory Moat

Definition

A law that makes it harder for new or small companies to compete, even if that wasn't the point — like a rule only a company with lots of lawyers can actually follow.

Super PAC

Definition

A group that can raise unlimited money to run political ads, as long as it doesn't coordinate directly with a candidate's campaign.

Duty of Care

Definition

A legal requirement to act carefully and responsibly — if you don't, you can be sued or punished, even if you didn't mean to cause harm.

Preemption

Definition

When a federal law cancels out state laws on the same topic, so states can't make their own — sometimes stricter — rules anymore.

The Players — And Why OpenAI and Anthropic Are Fighting Each Other

Same industry. Same regulatory question. Opposite money.

Team One National Rule

Led by OpenAI President Greg Brockman

Brockman personally gave $25 million to Leading the Future, a super PAC that has raised over $125 million and already spent $23.5-24 million on 2026 primary election ads. Its ads call state AI laws a 'chaotic patchwork' that would 'crush innovation.' The PAC spent $1.1 million attacking New York state legislator Alex Bores, author of New York's strict AI safety law, and $1.4 million backing four Texas primary candidates.

Team Let States Keep Their Own Laws

Backed by Anthropic

Anthropic gave $20 million to Public First Action, which has raised $20 million so far and spent $450,000 defending Bores against the Leading the Future attack ads. Anthropic publicly supports keeping state AI laws in place rather than one weaker federal standard — the opposite public position from OpenAI's own $25 million bet.

Team Rushed Safety Bill

Sens. Thune, Cruz, Klobuchar

Drafting a duty-of-care AI safety bill in direct response to a viral resignation post and Anthropic's own safety lead estimating a greater-than-10% chance of human extinction from AI within a decade. Being written now, with a real chance it doesn't happen this year — the House leaves for six weeks right after this window closes.

Team Anti-Moat

CREATE AI Act & GUARDRAILS Act sponsors

Two bills built to work against big-company advantage rather than for it: one hands government-funded compute access to university researchers and small startups; the other cancels the federal preemption order entirely so state laws survive. Neither has serious industry money behind it — which is itself the tell.

Team Nobody's Watching

AI PLAN Act, NO FAKES Act sponsors

Low-drama, low-lobbying bills — a fraud study requirement and a voice/likeness protection law — that pass quietly because nobody with deep pockets has a reason to fight them. OpenAI even publicly endorsed NO FAKES Act by name, a cheap way to look good with musicians while the real fight happens elsewhere.

Exhibit A · The Bill Everyone Is Racing Toward

Bill 1: The Thune-Cruz-Klobuchar AI Safety Bill

Three senators — Majority Leader John Thune, Ted Cruz, and Amy Klobuchar — are writing a duty-of-care law that would let the government say "no, you can't release that model, it's too risky" before it reaches your phone. It exists because an Anthropic employee named Jacob Coxon quit his job and warned publicly that AI companies are racing so fast they might accidentally build something lethal. Anthropic's own safety chief agreed with him.

OpenAI's lobbyist called Congress directly nine days before the recess window, while OpenAI publicly endorsed four California AI bills the same week

Confirmed

OpenAI's top lobbyist Chris Lehane personally called congressional offices on September 9, 2026, urging fast action. The same week, OpenAI publicly said yes to four California state AI laws right as the governor signed two of them — a calculated signal of cooperation timed to the federal push. No exact lobbying dollar figure has been disclosed for this specific bill yet.

Sources: IAPP, Technology.org, reporting on Chris Lehane outreach and California SB 813/AB 1405/SB 1119/AB 1864

The Ticking-Clock Tactic

A viral resignation post warning of human extinction hit 90 million views in 24 hours. Anthropic's own safety lead put a number on it: greater than 10% chance AI wipes out humanity within 10 years. That is textbook ticking-clock messaging — fear of imminent catastrophe pressures lawmakers to move fast and skip careful drafting, which usually means industry gets more input into the final text because there's no time for outside pushback.

The House leaves for a six-week election recess right after this week. If this bill doesn't move now, it likely doesn't move this year — which is exactly the deadline pressure that favors whoever is already in the room helping write it.

Estimated moat value: $1-$10 billion a year saved industry-wide if one national rule replaces 50 state rules. Regulatory moat: Yes, likely.

Exhibit B · The Real Fight, Fought With Campaign Ads Instead of a Bill

Bill 2: The State Moratorium / Preemption Fight

This isn't one bill — it's a fight over whether states keep the power to write their own AI safety laws. Congress tried to settle it twice and failed both times; once the Senate voted 99-1 against a federal preemption attempt. So the fight moved from the Senate floor into election spending — and it is, by dollar amount, the single biggest item on this entire tracker.

$125M+

Raised by Leading the Future (OpenAI-linked)

$23.5-24M

Already spent on 2026 primary ads

$20M

Raised by Public First Action (Anthropic-linked)

99-1

Senate vote against prior preemption bill

OpenAI is publicly supporting state laws while its president privately funds a $125M campaign to preempt them

Confirmed — Publicly Contradictory

Reporters at NPR, CNBC, and The Nation have specifically called out OpenAI for playing "both sides" — its president personally gave $25 million to a super PAC fighting to override state AI laws, while OpenAI's own public messaging supports state cooperation. The Nation notes the AI super PACs are directly copying the crypto industry's 2024 playbook: spend tens of millions on one issue, punish or reward candidates regardless of party.

Sources: NPR, CNBC, The Nation, WebProNews, Wikipedia (Leading the Future)

The Fear Appeal, Not the Merits

The attack ads call state AI laws a "chaotic patchwork" that would "crush innovation" — a jobs-and-competitiveness fear appeal, not an argument that the state laws are technically wrong. That framing is far easier to sell in a 30-second ad than a technical safety debate, and it borrows the same "don't let China win" logic used to justify less regulation across several other bills on this list.

Exhibit C · The Quiet Bill Riding Inside the Defense Budget

Bill 3: The GAIN AI Act

American companies, startups, and universities get first pick of Nvidia's newest AI chips before Nvidia can sell them to other countries, especially China. It has genuine bipartisan support — six senators from both parties introduced it together — and it already passed the Senate attached to a defense bill, which means it gets far less public scrutiny than a standalone vote would.

This bill mostly protects a spending advantage that already exists, and it moves through the NDAA process specifically because that draws less attention

Confirmed Pattern

Companies like OpenAI, Microsoft, Google, Meta, and Amazon already spend billions annually on chips — this bill just guarantees they get priority instead of competing globally for the same supply. Attaching it to the annual defense bill is the same well-worn move defense contractors have used for decades: bundle a provision into hundreds of pages that pass with minimal individual scrutiny.

Sources: Senate Banking Committee, Akin Gump legislative analysis

The Boomerang Risk

Past U.S. chip export bans already pushed China to pour billions into its own chip industry — Huawei, SMIC. Tightening the rules further risks accelerating exactly the Chinese self-sufficiency this bill is trying to prevent. Regulatory moat: Partial. Estimated value: $1-5 billion a year in protected advantage.

Exhibit D · The One Bill Built to Help the Little Guy

Bill 4: The CREATE AI Act

Right now only giant companies can afford the computing power to build cutting-edge AI. This bill creates a government-run lending library of compute — the National AI Research Resource — so university researchers, students, and small companies who could never afford their own supercomputer can borrow one through the National Science Foundation.

No major AI company is publicly lobbying for or against this bill — which is itself informative

Confirmed — Notable Absence

This is one of the only bills on the entire tracker that works against big companies staying on top rather than for it — it literally redistributes the resource that only incumbents currently have. Big AI companies have little reason to spend lobbying money helping their competitors catch up, and none appear to be doing so. Regulatory moat: No — anti-moat. It reduces incumbents' compute advantage by an estimated $100 million to $1 billion worth of access redistributed to smaller players.

Sources: Bill text (H.R.2385); absence of lobbying disclosure filings

Exhibit E-J · The Rest of the Tracker

Six More Bills, Ranked by Who They Actually Help

Bill	Plain-English Summary	Moat?	Before Nov?
5. AI PLAN Act (H.R.2152)	Passed the House. Just tells a federal agency to write a plan for stopping AI-driven financial fraud — no new rules, mostly homework.	Minimal	No
6. KIDS Act / SAFE BOTs Act (H.R.7757)	Passed the House in June 2026. Requires chatbots to disclose to minors they are talking to a bot, and bans manipulative design aimed at kids.	Mild	Maybe
7. NO FAKES Act (S.4591/H.R.8915)	Cleared Senate Judiciary in June 2026. Lets anyone sue over an AI-faked voice or likeness. OpenAI publicly endorsed it by name.	Barely any	No (27%/7% odds)
8. Data Center Moratorium Act (S.4214)	Long-shot bill freezing all new AI data center construction, citing Hinton and Bengio’s AI-pause warnings.	Not classic — freezes whoever is already built	No
9. Superintelligence Ban (Rep. Casar)	Announced Sept 15, 2026, no bill text yet. Would ban building superintelligence until a new oversight agency exists.	Same freeze-in-place caveat	No
10. GUARDRAILS Act "Right to Decide" (S.4216/H.R.8031)	Would cancel the December 2025 executive order and let states keep regulating AI. Backed by Anthropic’s public position.	No — anti-moat	No

The Pattern in the Quiet Bills

AI PLAN Act and NO FAKES Act move with almost no lobbying resistance because they cost everyone roughly the same, whether you're a two-person startup or a trillion-dollar lab. The Data Center Moratorium and Superintelligence Ban attract zero corporate lobbying against them for a different reason: openly fighting a "pause AI to keep us safe" bill looks terrible in the press right after a real AI safety scare — so companies that oppose them stay quiet rather than campaign publicly.

GUARDRAILS Act is the mirror image of the preemption fight in Exhibit B — same battle, fought in Congress instead of with campaign ads. Sen. Murphy's public statement warned AI "could destroy millions of jobs, addict kids to harmful chatbots, drive up utility bills, and deepen the crisis of loneliness" — a four-part fear appeal aimed at a completely different audience than the extinction-risk messaging elsewhere on this table, but built with the identical technique: pick the scariest available frame.

The Money, Side by Side

Bill	Estimated Moat Value / Stakes	Confirmed or Estimated?
1. AI Safety Bill	$1-$10B/yr industry-wide legal savings	Estimated
2. State Preemption Fight	$125M+ raised, $23.5-24M+ spent on ads	Confirmed — FEC filings
3. GAIN AI Act	$1-$5B/yr protected chip-access advantage	Estimated
4. CREATE AI Act	$100M-$1B in compute access redistributed away from incumbents	Estimated
5. AI PLAN Act	Under $50M industry-wide	Estimated
6. KIDS Act	$50-$300M relative advantage to large user-base platforms	Estimated
7. NO FAKES Act	Under $50M in company-side compliance cost	Estimated
8. Data Center Moratorium	$10-$50B+ (highest and least certain figure on the tracker)	Estimated
9. Superintelligence Ban	$10-$50B+, same reasoning as #8	Estimated
10. GUARDRAILS Act	~$0 in new advantage — preserves existing ~$1B/yr multi-state compliance cost	Estimated

The Historical Playbook, Repeating in Real Time

"AI super PACs are directly following the crypto industry's model."

— The Nation, on Leading the Future and Public First Action

Three precedents recur across this tracker, and none of them are subtle once you see the pattern:

The Russell-Einstein Manifesto model (1955)

Historical Precedent

Alarmed scientists' own warnings were used to build public pressure for arms-control talks that the nuclear powers then negotiated largely among themselves. A company that helps write the "scary" warning is well positioned to also help write the "solution" — the exact structure behind both the AI Safety Bill (Exhibit A) and the Superintelligence Ban.

The 2024 crypto super PAC playbook

Explicitly Copied

Spend tens of millions on one single issue, reward or punish candidates regardless of party, run attack ads that reframe the debate around economic fear rather than the substance. Leading the Future and Public First Action are both running this exact strategy, just for AI instead of crypto.

The NDAA attachment strategy

Confirmed Pattern

Attach a provision to the annual defense bill where it passes bundled with hundreds of other pages, far less scrutinized than a standalone vote. This is exactly how the GAIN AI Act is moving through Congress right now.

The One Argument That Shows Up in Almost Every Bill

The China Framing, Pointing in Both Directions

"Don't let China win" is used to justify less regulation in the preemption fight, the GAIN AI Act, and the safety bill's opposition — and to justify faster action in the same safety bill's supporters' framing. The Data Center Moratorium and Superintelligence Ban face the identical counter-argument in reverse: a US-only pause does nothing to stop a foreign lab from building the same thing first.

Only two bills on this tracker have essentially no China angle at all — the KIDS Act (aimed at chatbot companies regardless of where they're based) and the CREATE AI Act (which uses a mild pro-China-competition argument in the opposite direction: more total US teams racing beats a handful of giant companies racing alone). Critics note China already regulates its own AI industry, which undercuts the "any regulation helps China" framing used elsewhere — but that rebuttal rarely makes it into the 30-second ad version.

Case Status: Open

Confirmed

OpenAI’s Chris Lehane called Congress on Sept 9, 2026
Leading the Future raised $125M+, spent $23.5-24M+ on ads
Public First Action raised $20M, spent $450K defending Bores
GAIN AI Act passed the Senate attached to a defense bill
KIDS Act passed the House in June 2026
NO FAKES Act cleared Senate Judiciary; OpenAI publicly endorsed it
The Senate voted 99-1 against a prior federal preemption bill
Rep. Casar announced the Superintelligence Ban framework Sept 15, 2026

Estimated, Not Confirmed

Every dollar figure labeled "moat value" — modeled from public spending patterns, not disclosed by any company
Whether the AI Safety Bill passes before the House recess
The $10-50B+ figures for the Data Center Moratorium and Superintelligence Ban — untested, nothing like this has happened before

Read as Contradiction

OpenAI publicly supporting state AI laws while its president funds a $125M campaign to preempt them
"Chaotic patchwork" framing as an economic argument, not a merits-based one

The Punchline

Nobody Has to Be Lying

Jacob Coxon's fear is probably sincere. Anthropic's safety chief's estimate is probably a genuine belief. Sen. Murphy's warning about kids and loneliness is probably heartfelt. None of that is in question.

And it is still, simultaneously, true that OpenAI's president spent $25 million and Anthropic spent $20 million funding opposite sides of the exact fight this fear is being used to settle — with a combined $145 million already in motion before a single vote on the underlying preemption question, all inside a three-week window before the House leaves for six weeks.

Sincerity and convenience are not opposites. That is not a flaw in the story. That is the story.

References

[1] IAPP, "AI Safety Bill Tracker — Thune-Cruz-Klobuchar," iapp.org, September 2026.

[2] Technology.org, coverage of Senate AI safety bill drafting, September 2026.

[3] Semafor, "The rushed AI safety bill nobody wrote a draft of yet," September 2026.

[4] NPR, "OpenAI is funding both sides of the AI regulation fight," 2026.

[5] CNBC, "Leading the Future super PAC spending on AI primaries," 2026.

[6] The Nation, "AI super PACs are copying crypto's 2024 playbook," 2026.

[7] Wikipedia, "Leading the Future (political organization)," retrieved September 2026.

[8] FEC filings — Leading the Future ($125M+ raised, $23.5-24M spent); Public First Action ($20M raised).

[9] Senate Banking Committee, GAIN AI Act (S.3150/H.R.5885) markup records, 2026.

[10] Akin Gump, legislative analysis of GAIN AI Act and NDAA attachment strategy, 2026.

[11] CREATE AI Act (H.R.2385) — National AI Research Resource (NAIRR) program text.

[12] AI PLAN Act (H.R.2152) — House-passed text, National Science Foundation reporting requirement.

[13] KIDS Act / SAFE BOTs Act (H.R.7757) — House floor vote record, June 2026.

[14] Wikipedia and GovTrack, KIDS Act companion-chatbot minor-disclosure provisions.

[15] NO FAKES Act (S.4591/H.R.8915) — Senate Judiciary Committee vote, June 2026; GovTrack prognosis (27% Senate / 7% House).

[16] Office of Sen. Chris Coons, NO FAKES Act sponsor statement and OpenAI endorsement confirmation.

[17] Data Center Moratorium Act (S.4214) — bill text citing Hinton/Bengio open letter.

[18] Rep. Greg Casar, public announcement of Superintelligence Ban framework, September 15, 2026.

[19] GUARDRAILS Act ("Right to Decide") (S.4216/H.R.8031) — Rep. Beyer and Sen. Murphy sponsor offices.

[20] Office of Sen. Chris Murphy, public statement on AI job loss, chatbot addiction, utility bills, and loneliness.

[21] December 2025 Executive Order on AI federal preemption — White House text.

[22] U.S. Senate floor vote record, 99-1 rejection of prior federal AI preemption attempt.

[23] California SB 813, AB 1405, SB 1119, AB 1864 — state AI legislation, 2026 signing records.

[24] OpenAI public statement supporting four California AI bills, September 2026.

[25] Reporting on OpenAI lobbyist Chris Lehane's September 9, 2026 congressional outreach.

[26] Reporting on Jacob Coxon's resignation post and Anthropic safety lead's >10% extinction risk estimate.

[27] Russell-Einstein Manifesto (1955) — historical precedent for scientist-warning-driven policy.

[28] U.S. chip export policy analysis — Huawei/SMIC domestic chip investment response to prior export controls.

This report synthesizes publicly reported bill text, FEC filings, GovTrack prognosis data, and on-record statements available as of September 2026. Dollar figures labeled "estimated" or "moat value" are this report's own modeling based on public spending patterns and comparable past programs — they are not disclosed figures from any company or government agency unless specifically labeled "confirmed." The fear expressed by researchers, safety advocates, and lawmakers quoted here is presented as sincerely held. Where a claim is labeled Confirmed, it is independently verifiable through the cited sources. Sincerity and financial interest can coexist; this report describes that coexistence without asserting bad faith on the part of any named individual or organization.

6 views

0 likes

Start a Critical Discussion

These questions don't have consensus answers. Share one to LinkedIn or X and see what your network actually thinks.

"OpenAI's president personally gave $25 million to a super PAC fighting to kill state AI laws. Anthropic gave $20 million to a super PAC fighting to save them. Both companies are publicly saying something different from what their money is doing. Is that hypocrisy or just how a two-front war looks from the outside?"

"A viral resignation post hit 90 million views in 24 hours, citing a greater-than-10% extinction risk estimate — and it's being used to rush a safety bill through Congress in the same three-week window before the House leaves for six weeks. Is the urgency about the risk, or about the calendar?"

"Only two of 11 AI bills in Congress right now work against big-company advantage instead of for it — and neither has serious lobbying money behind it. Is the absence of money the most honest data point on the entire tracker?"

"The 'chaotic patchwork will crush innovation' attack ad doesn't argue state AI laws are wrong on the merits — it argues they're economically dangerous. Is that a policy argument or a 30-second ad script dressed as one?"

Share this analysis

If this changed how you think about something, share it. The AI workforce conversation needs more data and less hype.