Buist v. Anthropic: suing the labs for agreeing to the pause Hunt demanded in 2023

On 18 Sep 2026, four paid subscribers filed Buist v. Anthropic PBC, N.D. Cal. 3:26-cv-10693, against Anthropic PBC, OpenAI OpCo LLC, SpaceXAI LLC, and Google LLC. The claim is Sherman Act 1: a horizontal agreement to slow the rate at which competing frontier products improve, pled as output restriction, with Clayton Act injunctive relief and a jury demand. Lead counsel: Nick Rowley / Trial Lawyers for Justice; Andrew Tutt (Yale; former DOJ OLC). (complaint; Politico; Bloomberg Law.)

The public spine is six days old. 12 Sep: Dario Amodei, We Must Pace the Frontier ("We must slow the pace at which we improve the capabilities of AI models"; "industry-wide coordination"). Musk: "Dario is right." Altman: "I agree with Dario that we need to pace the frontier." Hassabis: "the right path forward," tying it to a July standards-body idea. 14 Sep: Altman says progress "should be slower than it otherwise could be" and OpenAI will not wait for an antitrust exemption. 15 Sep: OpenAI's Chris Lehane confirms weeks of talks with Anthropic and DeepMind. 18 Sep: complaint.

The predicate is older. The complaint treats the 28 Jul 2026 employee statement Pacing the Frontier (~1,178 signatories at OpenAI, Anthropic, Google, Meta, with organizational support from Encode AI and Guidelight AI Standards) plus a below-CEO working group since July as the conspiracy's machinery. Encode and Guidelight are therefore upstream of the facts the docket recites. They are not named as defendants.

The Hunt paradox is the reason this block exists. In April 2023, Cheyenne Hunt co-authored Public Citizen's Sorry in Advance!: "Until meaningful government safeguards are in place to protect the public from the harms of generative A.I., we need a pause." History, they wrote, gives "no reason to believe that corporations can self-regulate away the known risks." On 19 Sep 2026 she posted that she is suing because four labs agreeing to slow down is a "grift" - "rules written by the industry, for the industry, policed by the industry aren't safety standards" - and "the public should be writing the rules." The complaint's 7 is the same sentence in pleading form: guardrails "should not be decided by the Defendants, but by we the people, through state and federal regulation, and through the voices of juries."

That is not a flip. It is Public Citizen's 2023 line executed as an antitrust complaint: pause is legitimate if the state writes it; private coordination that preempts a statute is the thing to smash. The Sherman hook (subscribers get slower product improvement) is the standing theory. Hunt's own words are the framing.

1. What the complaint actually says (fact of the docket)

• Named plaintiffs: Charles Buist and Nick Spetsas (Florida; TL4J-adjacent trial lawyers; Hunt, Buist, Spetsas each allege paid subs to Claude, ChatGPT, Grok, and Gemini); Christine Bullock (California; Claude only). Ordinary class-action mix: two counsel-side names, one activist, one subscriber.
• Class: US persons who, from 12 Sep 2026, bought a paid individual sub directly from a defendant.
• Theory: competitors may each slow unilaterally; they may not agree to slow together so that "no participant bears the competitive cost of restraint" (5). Analog: carmakers do not need a cartel to refuse 300 mph passenger cars (9).
• They "take AI safety seriously" and are not against asking Congress for rules or even an exemption - they are against doing the coordination without one. Amodei himself flagged that a waiver would help. David Sacks had already called a waiver a "cartel." Hawley/Cruz blocked an NDAA antitrust carve-out days before the filing. Counsel executed a theory that was already live.
• No allegation that the safety concerns are fake.

A 29-page class complaint six days after the essay, with a July working-group narrative, is not four walk-in subscribers inventing Sherman Act doctrine on a weekend. That is a fact of the calendar, not proof of a hidden funder.

2. Hunt's paper trail (fact)

• Public Citizen Big Tech Accountability Advocate; co-author, Sorry in Advance! (18 Apr 2023), with Rick Claypool. Explicit pause-until-government-safeguards. Antitrust/self-preferencing already a chapter of that report.
• 2024 House campaign: AI/deepfake oversight, break-up-big-tech.
• 2026: founder of Reckoning Action (too new for a 990; donor list not public - absence is a fact). Swalwell/Platner misconduct work; WSJ profile. Klobuchar Judiciary clerk alumni channel. Ro Khanna as a political peer (Platner endorsement dropped on her advice; Khanna was talking "pace" with Chinese labs the day the complaint hit PACER, per The Information - same week, same noun, not a tasking proof).
• Gen Z for Change: Open Society Action Fund / Accelerate Action / Tides neighborhood. Progressive 501(c)(4) money, not Coefficient.

No public donation, retainer, or common counsel from Coefficient Giving (Open Philanthropy successor), GiveWell, FLI, CAIS, or Encode to Hunt, Reckoning Action, or TL4J turned up in open sources. That negative is part of the record.

3. The Encode / Guidelight hop (strongest indirect lead)

Pacing the Frontier (28 Jul 2026) is an employee letter asking the US government to help build tools that could later coordinate a slowdown. It is not a "pause now." Organizational support: Guidelight AI Standards (Page Hedley, Steven Adler - former OpenAI safety) and Encode AI. Signatories include Amodei, Kaplan, Pachocki, Chen, Hassabis-world safety leadership. Anthropic and OpenAI endorsed as companies.

The complaint recites that July statement as the moment defendants "recognized that no firm would slow alone" (49-52). Without Encode/Guidelight's midwifery, the docket's factual spine is much thinner.

Encode sits in the same responsible-tech / anti-preemption / "don't trust self-regulation" civil-society cluster as Public Citizen and Young People's Alliance (Omidyar Responsible Tech Youth Power Fund neighborhood). Hunt is a Public Citizen alumna. That is a two-hop: Encode helps produce the public event; a Public Citizen-trained advocate becomes the face of a suit that recasts that event as an illegal cartel and demands statutes instead.

Guidelight writes frontier safety standards meant to go beyond existing law. The complaint then treats the coordination those shops helped surface as the restraint of trade.

What this does not prove: Encode or Guidelight commissioned TL4J. If those shops "owned" Hunt, this filing is a strange order: it attacks private pacing and the waiver path, which is Public Citizen's line, not an EA antitrust carve-out.

4. EA / Coefficient / longtermism - map four money worlds, do not blob them

Effective Altruism is a real movement (GiveWell near-term vs longtermist x-risk split). Longtermism (Ord, MacAskill, Bostrom-adjacent) holds that future lives can dominate expected-value math, so existential-risk reduction can swamp present charity. Coefficient Giving is the 2024-26 rebrand of Open Philanthropy (Moskovitz/Tuna) - still the dominant outside funder of evaluators, alignment shops, and policy placement. This corpus already wires Open Phil -> OpenAI ($30M + board seat, 2017), EA -> Anthropic (LTBT), and EA -> the Nov-2023 OpenAI board crisis (catalog-nonprofits-foundations, catalog-elite-groups TESCREAL).

Critics (Gebru/Torres TESCREAL; the FTX/SBF scar, when Nick Beckstead ran Future Fund) argue longtermist EV can rationalize present harm and elite capture of AI policy. Some rationalist/EA houses have documented high-control dynamics. "Cult" and "turn yourself into a villain to maximize the future" are critic frames for a subset, not Coefficient's public grant theory and not a fact about Hunt. Grade them as contested ideology-critique, not as a finding that this complaint was so commissioned.

Beckstead now runs Secure AI Project (statutory safety protocols; state-bill circuit overlapping Encode). Howie Lempel (Coefficient/Open Phil alum) has been SAIP/Encode-neighborhood policy. That is personnel adjacency into Hunt's coalition space. It is not a check.

The incentive cut against capture: Coefficient-world wanted a lawful way to pace (standards body, embedded evaluators, Schiff-Banks-style waiver). This complaint attacks that public product as a Sherman conspiracy and makes the waiver harder. Coefficient did not need Hunt to file that.

a16z / Sacks lane: Sacks already called a pacing waiver "forming a cartel." The suit's legal theory is that sentence as a complaint. Hunt's press line is the opposite sermon (Congress must write the rules). Same filing, two uses. Overlap of interest is real. A funding trail is not.

Ethereum / x-risk donor set (Vitalik, Tallinn/SFF, historic overlap with Moskovitz and the SBF scar): no file on Reckoning Action, Hunt, or TL4J.

5. Classification (what an investigator can say)

Honest read: mixed vehicle. Legally packaged as subscriber harm from slower models. Publicly and personally used by its most visible plaintiff to advance a government-regulation agenda she has pushed since 2023, inside a progressive / responsible-tech neighborhood that sometimes touches Coefficient-adjacent people two hops out. The strongest indirect lead is Encode-Guidelight midwifery of the July letter, which the complaint then flips. That is association and alignment, not proven capture. Overlay. Not in the SCC.

6. How an influence operation actually looks here (not a single hand)

An operative does not need a check to Hunt. The usable architecture is predicate, frame, harvest.

Layer A - predicate factory (July). Encode AI and Guidelight AI Standards organize Pacing the Frontier (~1,178 employee signatures, 28 Jul 2026). The letter asks Washington to build the brake, not pull it. Anthropic and OpenAI endorse as companies. A below-CEO working group on a FINRA-style standards body is already meeting. The complaint later recites that letter as the moment defendants "recognized that no firm would slow alone."

Layer B - lab public coordination (12-15 Sep). Amodei's essay, same-day Musk/Altman/Hassabis assents, Altman "will not wait for an exemption," Lehane confirms weeks of talks. This is the overt act set.

Layer C - Washington frame already live (Jul-16 Sep). Schiff-Banks (S.5105, Collaboration on Adversarial Threats and Security Risks Act) would have given a narrow antitrust affirmative defense for safety coordination. 13 Sep: David Sacks (PCST chair; former White House AI czar; a16z-adjacent) tells the labs to slow unilaterally and to stop pretending they need a waiver "so you can form a cartel." 16 Sep: Cruz and Hawley reject the carve-out ("no world in which I consent"). Counsel did not invent Sherman-as-cartel. They harvested a frame that was already winning in the Senate and in Sacks's feed.

Layer D - harvest vehicle (18-19 Sep). TL4J files a 29-page class complaint in six days. Hunt's X post recodes the same facts as a "grift" and "we the people." Para 7's "voices of juries" is also TL4J's product: Rowley's career is MICRA-cap fights, Monsanto/Tesla verdicts, "lawyers who pursue money damages are civil-rights lawyers." A jury-as-regulator theory is what a national PI mill sells.

Layer E - dual-use of one filing. Hunt's sermon: statutes and juries, not lab-written rules. Sacks/Hawley sermon: no waiver, no cartel, product liability is enough. Same docket, two audiences. An accelerationist shop would like an injunction against lab coordination. A Public Citizen alum would like the clip that self-regulation is illegitimate. Neither needs to pay the other.

That is the op: not "EA ordered Hunt," but safety-network predicate + already-live cartel frame + PI mill + anti-self-reg advocate = one complaint that burns the waiver path the predicate's own donors wanted legalized.

7. Money and personnel on the predicate (open-source, dated)

Encode AI (Sneha Revanur). Public funder list: Responsible Technology Youth Power Fund (Omidyar neighborhood), Heising-Simons, Archewell, We Are Family, America's Promise, Future of Life Institute, Survival and Flourishing Fund, plus "rank-and-file" lab employees. Encode's own rule: no funding from corporations, foreign governments, or executives at frontier AI companies. That sentence is true as written and incomplete as economics.

SFF-2025 recommendations (Jaan Tallinn as funder): Encode AI Corporation, $516,000 general support (SFF-2025 recommendations). Tallinn is a Skype co-founder, a large x-risk philanthropist, and an early Anthropic investor. SFF-2024 also sent $700,000 general support to Longview Philanthropy. Encode GC Nathan Calvin was among accounts amplifying an Anthropic-adjacent "doom" post within minutes in Sep 2026, alongside Coefficient's Max Nadeau (press reconstruction; treat as reported timing, not a command net). Revanur has done the 80,000 Hours podcast (CEA's EA career org) - association, not a salary.

Guidelight. Founders: Page Hedley (OpenAI policy/ethics advisor; CSET fellow; AI program director at Longview Philanthropy; Williams & Connolly litigator; co-author of the "Not for Private Gain" letter against OpenAI's restructuring; co-author of an xAI safety report aimed at SpaceX investors) and Steven Adler (OpenAI 2020-24, product safety and dangerous-capability evals; discloses he still holds vested OpenAI equity). Guidelight says it accepts no funding from AI companies or their staff. Adler's equity is the disclosed exception. Hedley's Longview tour is the Coefficient-adjacent hop: Longview is an EA donor-advisor (Natalie Cargill) that has directed tens of millions in AIS grants and has been described as funded primarily by Coefficient; SFF's $700k is a second pipe. In August 2026 Guidelight's first control grades put Anthropic and OpenAI at C+, Google D+, xAI D-, Meta F - then the September complaint names SpaceXAI as a defendant on the back of Musk's "Dario is right." Do not over-read the grades as a targeting package; do record the sequence.

Coefficient Giving (Open Philanthropy rebrand, 18 Nov 2025): on track for ~$1B into catastrophic-risk / AIS in 2026; Navigating Transformative AI Fund; DC hiring for US AI policy (summer 2026); $160M to Resolution (Geoffrey Irving); placement culture is a critic claim with some alumni (Beckstead, Lempel) now at Secure AI Project, which has worked Encode's state-bill circuit (Utah HB 286 reporting). Coefficient did not need this lawsuit. The lawsuit makes the Schiff-Banks waiver harder. That is the anti-capture cut, and it still holds after the deeper dig.

Counsel layer. Andrew Tutt is on the caption at TL4J. Arnold & Porter (his former shop) advised D.E. Shaw Ventures on OpenAI Series C and Anthropic Series G/H in 2026 - the opposite side of the table. That cuts against "the lawyers are a lab front" and also against "the lawyers are an EA front." Rowley is a $3.5B-verdict PI entrepreneur who already sues Tesla. Contingency economics explain the vehicle without a Coefficient retainer.

8. Targeting cycle (calendar as evidence)

Six days from essay to PACER is fast because the theory and the exhibits were already in public. That is tradecraft of a harvest, not proof of a hidden principal.

9. Cui bono (who can use the filing without writing the check)

The people who look like they midwifed the predicate are the ones the complaint hurts. The people who look like they benefit from the cartel frame (Sacks, Hawley, a PI mill, a Public Citizen alum) have no documented payment to each other. That is what a flip looks like.

10. What a professional investigator still cannot say

• No bank trail, 990, retainer, or common counsel from Coefficient, SFF, FLI, Encode, or Guidelight to Hunt, Reckoning Action, Buist, Spetsas, Bullock, or TL4J.
• No file that Encode "ordered" the flip of its own letter.
• Reckoning Action donors remain invisible (org too new). That absence is a fact, not a Coefficient check.
• Khanna talking "pace" with Chinese labs the day of PACER is same-week, same noun - not tasking.
• Adler's OpenAI equity and Hedley's Longview tour are disclosed biography, not kompromat.
• Encode's no-executives rule vs Tallinn/SFF is a substance-over-form point, not a lie in the sentence they published.

Updated classification: the operation is real as architecture (predicate factory -> live cartel frame -> harvest complaint). It is not proven as command (Coefficient/EA/Encode directing Hunt). Hunt is using a novel antitrust hook to advance a 2023 government-rules agenda. The safety network built the event she flipped. Accelerationist Washington supplied the word "cartel" five days earlier. TL4J supplied the jury machine. Overlay. Not in the SCC.

What is NOT asserted

• The labs' safety concerns are fake (the complaint does not say that; neither do we).
• A secret EA/Coefficient handler wrote the complaint.
• Hunt is an EA agent. Her 2023 pause and 2026 suit are the same anti-self-regulation demand.
• Khanna tasked the filing because he was talking "pace" that week.
• Longtermism as a whole is a cult, or that "short-term suffering is the point." Describe the EV-domination critique; do not flatten GiveWell malaria nets into SBF.

Sources: Complaint, 3:26-cv-10693; Hunt, 19 Sep 2026; Public Citizen, Sorry in Advance!, 18 Apr 2023; Forbes on Pacing the Frontier + Encode/Guidelight; Encode who-we-are / funders; SFF-2025 Encode $516k; Guidelight about/team; Sacks cartel line; Cruz/Hawley reject carve-out; Arnold & Porter / D.E. Shaw OpenAI+Anthropic; Coefficient $1B AIS 2026. Cross-refs: OpenAI, Anthropic, Google, xAI, SpaceX, Effective_Altruism, Open_Philanthropy, Coefficient_Giving, Encode_AI, Guidelight, Public_Citizen, a16z, TESCREAL, Dustin_Moskovitz, Survival_and_Flourishing_Fund, Longview_Philanthropy, Jaan_Tallinn, David_Sacks.

← Research index · structured data: spec-buist-v-anthropic-pacing.json · spec-buist-v-anthropic-pacing.md