A reading of one week in September 2026: a bill to ban superintelligence, two frontier model launches on the same day, and a United Nations that wants a brake it cannot build on its own. The question running through all of it is the oldest one in accountability reporting. Who benefits?
On Wednesday, September 23, 2026, Senator Bernie Sanders and Representative Greg Casar unveiled the Ban Artificial Superintelligence Act. The bill would permanently prohibit the development and deployment of AI that exceeds human cognitive performance across most domains or could destroy or disempower humanity, pause advanced AI development until a new federal Department of Artificial Intelligence has written safety rules, require companies to obtain approval before deploying advanced systems, and carry prison terms of up to 20 years for violations (PBS News, 2026; Roll Call, 2026). It also directs the United States to pursue international agreements to stop superintelligence being built anywhere. On the same Wednesday, according to The Register, Anthropic released Claude Opus 5.5 and OpenAI released GPT-6 Sol and Luna, continuing a release cadence that has moved from roughly quarterly in 2025 to almost monthly in 2026 (Claburn, 2026). And across town, world leaders were arriving for a General Assembly high-level week at which the Secretary-General had named runaway AI as one of three existential threats facing the world (United Nations, 2026a).
Three events in one week are not a conspiracy. They are a snapshot, and a revealing one. Each actor in that snapshot is saying some version of the same thing, that frontier AI may need to slow down, while doing something that serves a different interest. My job here is not to tell you who is right about superintelligence. It is to follow the documents, the money and the timing, and to show you where the stated goals and the incentives line up and where they part company.
What the bill actually does, and who it helps
Start with the text as reported, because the headlines flatten it. The ban itself targets a capability that, by the bill’s own definition, does not yet exist. The operative near-term provision is the pause on advanced AI development until a federal regulator has written rules, followed by a pre-approval regime in which firms submit development plans and seek sign-off before deployment (Roll Call, 2026). Roll Call reports that the draft identifies precursor characteristics, such as the capacity to uplift the design of nuclear, chemical or biological weapons, to deceive humans or to modify its own functions, which is where any real enforcement would bite.
The bill’s most immediate beneficiary is political, and there is nothing cynical in saying so. PBS reports that the bill faces long odds in a Republican-controlled Congress and sits within a progressive strategy ahead of the midterms and the 2028 cycle (PBS News, 2026). Sanders has framed it as a fight against AI oligarchs and against a campaign finance system that the industry is using to pour tens of millions of dollars into this year’s elections (Common Dreams, 2026). A bill that will not pass can still do work: it defines the far end of the debate, gives candidates a position to be for or against, and forces the industry’s political operations to reveal what they will pay to oppose.
The less obvious question is who would benefit if a pre-approval regime like this ever did become law. Licensing regimes tend to favor the firms that can afford the compliance teams, the legal departments and the relationships with the regulator. A mandatory approval process for advanced models would raise the cost of entry for smaller labs and open-weight developers far more than for the handful of companies that already employ policy staff by the hundred. That is not an argument against the bill; it is a design risk its drafters would need to answer, and it is the argument its industry critics will reach for first. The Alliance for Secure AI, which welcomed the bill, noted that some in the industry object specifically to a domestic pause without an international agreement (Roll Call, 2026), which is the objection that keeps the race framing alive.
The employees, the letter and the release calendar
The most striking detail in the coverage is who endorsed the bill. PBS reports that current employees of OpenAI and Google DeepMind put their names to a statement supporting it, including an OpenAI safety systems researcher and a DeepMind engineer who warned that the current path leads toward catastrophe (PBS News, 2026). That builds on Pacing the Frontier, a statement published in July and now listing 1,386 signatories from frontier companies, including Anthropic’s chief executive, OpenAI’s chief scientist and a co-founder of Google DeepMind (Pacing the Frontier, 2026). Its request is careful. It asks Washington to support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development. It does not ask for a pause.
Here the incentives are worth sitting with. Both OpenAI and Anthropic endorsed that letter as companies within a day of its publication, according to multiple reports, and Anthropic’s chief executive reiterated its message in a post on September 12 (Claburn, 2026). Eleven days later both companies shipped new flagship models on the same day. I do not read that as hypocrisy so much as a precise statement of position: an option to pace, held internationally, costs a company nothing today and protects it against being the only one that slows down tomorrow. A unilateral pause, of the kind the Sanders bill would impose on American firms, is exactly what the letter was written to avoid. The same season produced two slowdown proposals, one that the leading labs signed and one that they did not, and the difference between them is who bears the cost of going first.
Follow the money
The political money makes the split visible. Leading the Future, the super PAC launched with $125 million in announced commitments, had raised $75.1 million by the end of the first quarter, according to Public Citizen’s analysis of Federal Election Commission filings, with $50 million from Andreessen Horowitz and $25 million from OpenAI president Greg Brockman and his wife (Public Citizen, 2026). Its stated purpose is to grow pro-AI candidates and an innovation agenda, and it spent heavily against a New York legislator who sponsored the RAISE Act on AI safety (Public Citizen, 2026). On the other side, Anthropic gave $20 million in February to Public First Action, a 501(c)(4) organization that supports lawmakers favoring AI safety rules, and later doubled that commitment to $40 million; the company has said its money cannot be used to influence candidates, while an affiliated super PAC does the campaigning (CNBC, 2026).
Read those filings together and a pattern emerges that the ban debate obscures. The industry’s political spending is not a contest between building and stopping. It is a contest between two kinds of builders over who writes the rules: those who want a light national framework that preempts the states, and those who want mandatory safety testing and room for state oversight. Neither camp, as far as the public record shows, is funding candidates because they support a ban. The Sanders bill is, in that sense, the one position in the room that no frontier lab is paying for, which is either its weakness or its point.
What the United Nations can and cannot do
At the United Nations, the Secretary-General has been unusually direct. Ahead of the high-level week he said the world cannot afford a race to the bottom on AI safety, that voluntary pauses will not work if confined to isolated countries, and that global cooperation is indispensable (United Nations, 2026a). He has also argued that if AI development should slow when risks become too great, good intentions are not enough, and that the countries with the highest capacity should establish mechanisms of contact to avoid a race to the bottom (United Nations, 2026b). The UN now has two instruments created in August 2025: an Independent International Scientific Panel on AI, which issued a preliminary report on July 1, and a Global Dialogue on AI Governance, which met for the first time on July 6 and 7 and will meet again in New York on May 3 and 4, 2027 (United Nations, 2026b; United Nations, n.d.).
Who benefits from the UN track? Middle powers and the Global South gain a forum in which they have a voice on a technology built almost entirely in two countries. The frontier labs gain something too, because an international effort is exactly what Pacing the Frontier asked for, and a process that meets once a year creates no binding obligation on anyone. What the UN lacks is the thing the Secretary-General keeps pointing at: a mechanism among the handful of states that actually host frontier development. Without Washington and Beijing at its center, the Global Dialogue can describe a brake but cannot install one.
The pattern
Put the week together and each actor’s position tracks its exposure. The labs support pacing that is international, optional and future-facing, because it limits the risk of being overtaken without constraining what they ship this quarter. The accelerationist money supports national preemption, because it removes the state legislatures that have been the fastest source of binding rules. The safety-aligned money supports testing mandates, which well-resourced safety-focused firms can meet more easily than their competitors. The Sanders bill supports the one outcome none of them want, a domestic stop, and its sponsors gain a clear line in a campaign season. The UN supports cooperation it cannot compel. None of this tells you whether superintelligence is near or whether a ban is wise. It tells you to read every proposal in this debate by asking who pays first, and who would still be standing if it passed.
Disclosure: this article was researched, written and reviewed by FairByDesign. Claude, an AI model made by Anthropic, one of the companies discussed, was used for editing. FairByDesign accepts no funding from any AI developer named here.
References
Claburn, T. (2026, September 23). Frontier AI keeps racing despite calls to slow down. The Register. https://www.theregister.com/ai-and-ml/2026/09/23/frontier-ai-keeps-racing-despite-calls-to-slow-down/5298448
CNBC. (2026, February 12). Anthropic gives $20 million to group pushing for AI regulations ahead of 2026 elections. https://www.cnbc.com/2026/02/12/anthropic-gives-20-million-to-group-pushing-for-ai-regulations-.html
Common Dreams. (2026, September 23). Sanders, Casar unveil bill to ban AI superintelligence that “could kill countless numbers of people.” https://www.commondreams.org/news/sanders-casar-superintelligence-ban
Pacing the Frontier. (2026). Pacing the frontier [Joint statement]. Retrieved September 28, 2026, from https://www.pacingthefrontier.com/
PBS News. (2026, September 23). Sen. Bernie Sanders unveils bill to ban artificial superintelligence and create Department of AI. https://www.pbs.org/newshour/politics/sen-bernie-sanders-unveils-bill-to-ban-artificial-superintelligence-and-create-department-of-ai
Public Citizen. (2026, June 30). Corporate supremacist super PACs drive $500 million midterm spending surge. https://www.citizen.org/article/corporate-supremacist-super-pacs-500-million-midterm-crypto-ai-sports-betting/
Roll Call. (2026, September 23). AI “superintelligence” ban proposed by Casar, Sanders. https://rollcall.com/2026/09/23/ai-superintelligence-ban-proposed-by-casar-sanders/
United Nations. (2026a, September 16). AI, climate and conflicts top Guterres’s agenda ahead of General Assembly. UN News. https://news.un.org/en/story/2026/09/1168348
United Nations. (2026b, September 16). Who should set the rules for AI? The UN is pushing for a safer digital future. UN News. https://news.un.org/en/story/2026/09/1168353
United Nations. (n.d.). Global Dialogue on AI Governance. Retrieved October 9, 2026, from https://www.un.org/global-dialogue-ai-governance/en
Leave a Reply
You must be logged in to post a comment.