On March 25, 2026, Sen. Bernie Sanders (I-VT) and Rep. Alexandria Ocasio-Cortez (D-NY) introduced a bill that would do something no state legislature has attempted: freeze all new AI data-center construction nationwide, with no fixed end date, until Congress passes federal legislation covering AI safety review, worker protections, environmental safeguards, and civil-rights protections.[1] Five months later and on the opposite side of the argument, on the same September week that Anthropic's Dario Amodei asked government to consider "mediation or waivers of antitrust restrictions" so frontier labs could jointly pace their own development, former FTC Chair Lina Khan posted the opposite case: AI companies do not need a new legal exemption, because "there's no AI exemption from laws already on the books."[2] Different targets, different legal levers, same industry, same season.
S.4214, the Artificial Intelligence Data Center Moratorium Act, was referred to the Senate Commerce Committee the day it was introduced and has not advanced since.[3] Its structure is blunt: an immediate, nationwide construction and upgrade freeze on AI data centers that only lifts once Congress enacts a law meeting a detailed list of conditions -- prerelease federal safety and effectiveness review for AI products, worker protections, consumer protections, environmental safeguards, and civil-rights protections -- plus new restrictions on exporting AI computing hardware to countries without comparable laws.[1] "We cannot sit back and allow a handful of billionaire Big Tech oligarchs to make decisions that will reshape our economy, our democracy and the future of humanity," Sanders said in the announcement. Ocasio-Cortez put the same argument in constitutional terms: "Congress has a moral obligation to stand with the American people and stop the expansion of these data centers until we have a framework to adequately address the existential harm AI poses to our society."[4]
This outlet has already covered twelve states' worth of data-center restriction bills, a Texas governor's ERCOT queue freeze, and Ohio and Maine's moratorium fights.[5] Every one of those was local or state action, and every one of them was triggered by a physical or fiscal constraint close to the ground: grid capacity, noise, water draw, lost tax revenue. S.4214 is a different species of objection entirely -- federal, motivated by AI-safety and labor concerns rather than grid math, and written by the most visible standard-bearers of the Democratic Party's left flank rather than a county board or a Republican governor auditing a queue.
Stated plainly, because the honest version matters more than the dramatic one: S.4214 is a proposal, not law. It was referred to committee in March and has seen no committee vote in the nearly six months since.[3] It was introduced into a Congress where both chambers are controlled by Republicans who have, as of this month, had the sitting president personally reject the idea of slowing AI development at all -- "whoever wins AI wins," as Trump put it two days before this piece was published, in direct response to the industry's own, much more modest pacing proposal.[6] If the White House will not entertain a voluntary industry slowdown, a mandatory federal construction freeze written by Sanders and Ocasio-Cortez has essentially no near-term path to a floor vote, let alone passage. That does not make the bill meaningless -- introduced legislation shapes the boundaries of what gets discussed even when it does not pass -- but it is a marker being planted, not a rule being enacted.
The second half of this story runs on a different legal track. Post 417 on this outlet already documented that Anthropic, OpenAI, and Google representatives had been meeting since July 2026 to discuss a joint AI industry standards body, and that Amodei's own September 12 essay -- the public event that made that private coordination legible -- explicitly called for pacing coordination "with government mediation or waivers of antitrust restrictions."[7] Competitors agreeing to jointly slow their own output is close to a textbook antitrust problem, which is exactly why Amodei asked for cover rather than just doing it.
On September 14, 2026 -- the same week -- Lina Khan, who ran the FTC from 2021 to early 2025 and has been one of the sharpest institutional voices on Big Tech concentration since, posted publicly that the industry does not need any such waiver. "Law enforcers already have authority to charge companies and their CEOs for creating and releasing dangerous, unvetted, or defective products," Khan wrote. "We shouldn't let discussions about new legal regimes distract from the fact that there's no AI exemption from laws already on the books."[2] She was specific about the mechanism: shipping a flawed AI product without adequate safeguards can already qualify as an "unfair or deceptive" practice under the FTC Act, and using competitively sensitive customer information across an interconnected web of AI partnerships and cross-investments is already the kind of conduct existing "unfair methods of competition" law was built to catch.[2] The available reporting on Khan's post does not show her naming Amodei or the standards-body meetings directly -- her framing is general, aimed at "new legal regimes" for AI broadly -- but the timing and substance land squarely against what post 417 already documented the labs are asking for. Read next to each other, Khan is not describing a hypothetical: she is describing the same industry, the same week, making the opposite argument to the one its own most safety-forward CEO just made in public.
The Sanders/Khan pairing is not the only place this argument surfaced this week. Antitrust scholar Dirk Auer published a piece on Truth on the Market, also dated September 14, 2026, titled "Move Slow and Collude: The Antitrust Problem With Pacing AI," arguing the legal case in more technical terms than either Sanders or Khan attempted.[8] Auer's core claim: "a collective agreement among frontier labs to pace themselves...looks like a textbook cartel" -- an arrangement that restricts output and capability growth in exactly the way antitrust law was written to prohibit, regardless of the safety rationale behind it. His sharper point is that safety and pacing are not the same lever: independent evaluators embedded inside labs (the part of Amodei's plan Auer supports) do not require competitors to coordinate with each other at all, while a shared industry ceiling on capability growth does -- and "a pacing cartel could therefore slow AI development without making AI safer."[8] Auer is a position, not a ruling, same as Khan -- but it is an independently argued position from a specialist publication, not a restatement of Khan's framing, and it reaches the same conclusion from a different direction: the antitrust waiver the industry is asking for is doing real legal work, not just clearing a formality.
Three different objections, three different legal levers, converging on one industry at the same moment it is asking for more legal room, not less. Post 425 documented state and local pushback -- Virginia's rate rules, Loudoun County's noise ordinance, Ohio's tax-exemption pause, Mendocino's emergency ban -- nearly all of it triggered by a physical or fiscal constraint close to the ground: grid capacity, water, noise, lost revenue. Sanders and Ocasio-Cortez's bill is a federal version of restriction, but built on an entirely different foundation: not grid math, but AI-safety and labor politics, aimed at Congress rather than a county board. Khan and Auer are a third category again -- neither proposing a new restriction nor blocking a permit, but arguing the restrictions AI companies already face under existing antitrust and consumer-protection law are being underused, not that new ones are needed. None of these three actors coordinated with each other. They arrived at overlapping pressure on the same industry, in the same season, from the political left, from federal legislators, and from antitrust law itself -- while the industry's own request, documented in post 417, runs the opposite direction: less legal exposure, not more.
Why does this matter? None of this changes the AI industry's actual legal exposure today. S.4214 has no realistic path through a Republican Congress that just watched its own president reject a voluntary industry slowdown; Khan is a former official making a public argument, not a sitting regulator issuing an order; and Auer's antitrust analysis is a scholar's reading of the law, not a court's. What is real is that a federal bill exists, with real sponsors and real text, going further than any state or county restriction this outlet has covered. What is real is that a former FTC chair with a well-documented record on Big Tech concentration put her name on the claim that the industry's own antitrust-waiver request is unnecessary. And what is real is that the industry is asking for expanded legal room -- a standards body, an antitrust waiver -- at the exact moment three separate, ideologically distinct sources of pressure are arguing for less of it, not more. The gap between a proposal and a law is real and worth stating plainly. So is the gap between an industry asking for cover and the growing list of people telling it, from three different directions, that it does not need any.
Companion pieces on this outlet: "Anthropic's CEO Called on the AI Industry to Deliberately Slow Itself Down...While a Ratings Agency Modeled, Not Predicted, a 35% Stock Crash...", which first documented the industry's antitrust-waiver request this piece answers, and "China's Ulanqab Hub Generates 848 Million Kilowatt-Hours a Year to Skip the Grid Queue Entirely. Twelve U.S. States Introduced Bills in 2026 to Slow Their Own Data Centers Down -- On Purpose.", which documented every state and local restriction this federal bill goes further than.