Who Benefits When Everyone Says “Slow Down”?

In one September week, a US senator proposed banning superintelligence, two frontier labs shipped new flagship models on the same day, and the UN Security Council heard that AI could slip beyond human control. Everybody, it seemed, wanted to slow down. This is the opening of a dialogue about the question that week kept skipping: who benefits from each version of “slow”, and who gets left out of deciding?

Here is a sentence that should make you sit up. On 12 September 2026, Anthropic’s chief executive, Dario Amodei, wrote that “we must slow the pace at which we improve the capabilities of AI models,” and eleven days later his company released a new flagship model, on the same day OpenAI released two of its own (Claburn, 2026). I am not quoting that to score a point. I am quoting it because it is the cleanest illustration I have found of the problem this series exists to examine. The people saying “slow down” and the people deciding how fast things go are, very often, the same people. And the rest of us, wherever we live and whatever we do for a living, are mostly watching.

So let me ask you to do something before you read on. Think about the AI system that most affects your work, your studies or your data right now. Then ask who decided how it was built, what it was allowed to do, and what would happen if it went wrong. If your honest answer is “I don’t know, and nobody asked me,” you are the person this dialogue is for.

One week, four kinds of “slow”

Start with the documents, because headlines flatten them. On 23 September, Senator Bernie Sanders and Representative Greg Casar unveiled the Ban Artificial Superintelligence Act. As reported by PBS News and Roll Call, it would permanently prohibit AI that exceeds human cognitive ability or could destroy or disempower humanity, pause advanced AI development until a new federal Department of Artificial Intelligence has written safety rules, require approval before advanced systems are deployed, ban systems with “precursor” characteristics such as the capacity to uplift weapons design, deceive humans or modify themselves, and carry prison terms of up to 20 years in some cases (PBS News, 2026; Roll Call, 2026). Sanders put the case in one line: “It doesn’t take a genius to say, ‘slow it down’” (PBS News, 2026). PBS also reported that the bill faces long odds in a Republican-controlled Congress. In the House, Casar introduced it on 24 September as H.R. 10538, with a group of House cosponsors, and it was referred to the Committee on Oversight and Government Reform (U.S. Government Publishing Office, 2026).

The second “slow” is the one the industry itself signed. Pacing the Frontier, a statement published in July 2026, now lists 1,386 signatories from AI companies, including Amodei, OpenAI’s chief scientist Jakub Pachocki and Google DeepMind co-founder Shane Legg. Its request is careful: that the US government “support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development” (Pacing the Frontier, 2026). Notice what it does not ask for. It does not ask for a pause, and it does not ask any one company to go first.

The third “slow” lives at the United Nations. The day the bill was unveiled, the Security Council was briefed by OpenAI’s chief executive Sam Altman and by Yoshua Bengio, who warned that uncontrolled AI is “an unprecedented threat for countries” that none can contain alone (United Nations, 2026c). Two days earlier, the UN’s Independent International Scientific Panel on AI had published a thematic brief on misalignment risks from AI agents, in which Bengio, a co-chair, said “the traditional model of safeguarding is unravelling” (United Nations, 2026d). The Secretary-General’s own framing has been consistent: “If we believe AI development should slow when risks become too great, we need more than good intentions” (United Nations, 2026a).

The fourth “slow” is a “don’t”. At the General Assembly debate, the United States president said, “I’m not going to stifle growth of something that will be bigger than the Industrial Revolution” (United Nations, 2026b). That position has money behind it. Leading the Future, a super PAC whose stated aim is “identifying, maintaining, and growing pro-AI candidates,” had raised $75.1 million by mid-2026, according to Public Citizen’s analysis of federal filings, including $50 million from Andreessen Horowitz and $25 million from OpenAI president Greg Brockman and his wife (Public Citizen, 2026). On the other side of the ledger, Anthropic gave $20 million in February 2026 to Public First Action, a bipartisan 501(c)(4) group that campaigns for AI safeguards, and a further $20 million in July, bringing its total to $40 million; the company says both gifts are restricted to the group’s public education and policy work and cannot be used to influence the election of any candidate (Anthropic, 2026a, 2026b).

Reading each “slow” by who pays first

What follows is FairByDesign analysis, not reporting of anyone’s admitted motive. Read the four positions by asking who bears the cost if each one wins, and a pattern appears. The labs’ version of slowing down is international, optional and future-facing: it costs nobody anything this quarter and protects every signatory from being the only one that stops. The Sanders version is domestic and immediate, which is precisely why no frontier lab is funding it. The deregulatory version puts the cost on whoever is harmed later, and on the state legislatures that had been the fastest source of binding rules. The UN version asks for cooperation it cannot compel; as the draft that preceded this piece put it, a dialogue that meets once a year can describe a brake but cannot install one.

None of this tells you whether superintelligence is near or whether a ban is wise. It tells you something more useful. Every proposal in this debate distributes costs and benefits, and the people most able to shape the proposals are the ones with the most to gain or lose from them. That is not a conspiracy. It is how power usually works when a technology is concentrated in a handful of firms and a couple of countries.

The people who were not in the room

Now look at who else spoke that fortnight, because they told a different story. At the high-level debate, South Africa tied inclusion in rule-making to “research, youth and economic transformation”; India described AI as a driver “for human empowerment, economic prosperity and population-scale problem solving”; China called on states to “narrow the digital divide”; and France warned of dependence “on either dominant States or a handful of private firms” (United Nations, 2026b). Singapore floated a future UN framework convention and a dedicated international AI institution (United Nations, 2026b). These are not the same positions, and some of these governments have their own reasons to want a seat. But they share a complaint: the rules are being drafted somewhere else.

The numbers behind that complaint are stark. When the UN’s High-level Advisory Body on AI mapped seven major international AI governance initiatives in 2024, it found that seven countries were party to all of them and 118 countries, mainly in the Global South, were party to none (United Nations High-level Advisory Body on Artificial Intelligence, 2024). The body that was created partly to fix this, the Global Dialogue on AI Governance, held its first session in Geneva on 6 and 7 July 2026 and meets next in New York on 3 and 4 May 2027 (United Nations, n.d.). Its own description promises governance that reflects “the priorities of all nations, not just the most technologically advanced.” Whether it delivers is an open question this series will keep asking.

Then there were the voices that rarely make the briefing note at all. At a UN meeting on children and AI on 22 September, Volker Türk, the UN High Commissioner for Human Rights, observed that “today’s children are the first generation to grow up with artificial intelligence – yet often the last to be asked for their views,” and UN News reported on a survey of more than 1,000 children in 49 countries whose concerns included deepfakes, data exploitation and eroded creativity (United Nations, 2026e). Two weeks later, in Oxford, Türk went further: “We can no longer afford to wait or rely on goodwill,” he said, calling for mandatory human-rights due diligence and a bigger role for civil society in AI decisions (United Nations, 2026f). And Girmaw Abebe Tadesse, a member of the UN Scientific Panel who works in Kenya, made the point that should frame any conversation about “slowing down” from a continent that has barely been allowed to speed up: “It is important to ask ourselves what problems we want to solve first” (United Nations, 2026g).

Why this is a dialogue, not a verdict

I could end here with a neat conclusion. I don’t think I should. FairByDesign investigates the complex middle where AI decisions become systems, and the honest position on “who benefits from slowing down?” is that it depends on whose slowing, on what terms and enforced by whom. That is exactly the kind of question that gets settled by default when only the beneficiaries are talking. The editor’s instruction for this strand is blunt, and I agree with it: the point is to bring academic, technical, operational, research, policy and societal voices from across cultures into the same conversation, so that people have transparency and a say in their future, rather than leaving decisions to the privileged beneficiaries of the technology revolution.

So here is the plan. Over the next four instalments, every second Wednesday, Who Benefits takes one “who decides, who benefits, who carries the cost” question and puts it in front of people who see it from different places. WB-2 asks who writes the technical standards that turn AI laws into practice, and who can afford a seat at that table. WB-3 follows AI infrastructure to the towns that host it, where the electricity, the water and the land are local but the profits are not. WB-4 looks at young people and entry-level work, and at who gets to decide how that transition is managed. WB-5 asks whose languages and whose knowledge train these systems, and what communities in the Global South are building on their own terms. Each piece will use verified sources and real, attributed positions, and each one will end by asking you in.

If you are young, this matters in a particular way. You are not too early to understand this, and you are not too young to question it. The decisions are being made now, and you are entitled to see them while they are being made, not after.

Disclosure: this article was researched, written and reviewed by FairByDesign. Claude, an AI model made by Anthropic, a company active in the debates discussed here, was used for editing. FairByDesign is solely responsible for its content. FairByDesign accepts no funding from any AI developer named here.

Join the dialogue

This series only works if it hears from people who are not usually asked. We especially want to hear from researchers and practitioners working outside the usual capitals, in Africa, Asia, Latin America, the Middle East and the Pacific; from engineers and safety staff inside AI companies speaking in a personal capacity; from policy staff in middle-power governments and regional bodies; from civil-society, labour and youth organisations; and from students who study or organise around technology governance. Tell us which version of “slow” you think serves the public, which one you think serves the incumbents, and what evidence we have missed.

Send evidence, corrections or a serious disagreement to info@fairbydesign.org with “Who Benefits” in the subject line; for sensitive material, see how we handle sources. With your permission, responses may be quoted in a follow-up post or become a recorded conversation on the podcast. And you can give without money too: share this piece, suggest a lead, introduce us to someone we should hear from, or tell us your story. All responses to the Who Benefits dialogue go to info@fairbydesign.org.

Support the dialogue. Cross-cultural conversations like this one don’t happen by default; they happen when someone has the time to find the voices that the people with the most to gain would rather not hear. Give once ($5, $10, $25, $50 or any amount), give monthly ($10 or $25), or become a patron at fairbydesign.org/patrons-2. Support never buys editorial control.

References

Anthropic. (2026a, February 12). Anthropic is donating $20 million to Public First Action. https://www.anthropic.com/news/donate-public-first-action

Anthropic. (2026b, July 21). Donating another $20 million to Public First Action. https://www.anthropic.com/news/donation-public-first-action

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

Pacing the Frontier. (2026). Pacing the frontier [Joint statement]. Retrieved October 7, 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/

U.S. Government Publishing Office. (2026). H.R. 10538 (IH): Ban Artificial Superintelligence Act of 2026. 119th Congress, 2nd session. https://www.govinfo.gov/app/details/BILLS-119hr10538ih

United Nations. (n.d.). Global Dialogue on AI Governance. Retrieved October 7, 2026, from https://www.un.org/global-dialogue-ai-governance/en

United Nations. (2026a, 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. (2026b, September 30). Who gets to shape AI? UN debate centres on power, trust and inclusion. UN News. https://news.un.org/en/story/2026/09/1168498

United Nations. (2026c, September 23). OpenAI and Anthropic brief Security Council amid “real and imminent” threat posed by runaway AI. UN News. https://news.un.org/en/story/2026/09/1168414

United Nations. (2026d, September 21). UN panel calls for stronger safeguards as AI agents advance. UN News. https://news.un.org/en/story/2026/09/1168380

United Nations. (2026e, September 22). “Ask us first”: Children demand a voice on AI’s future. UN News. https://news.un.org/en/story/2026/09/1168403

United Nations. (2026f, October 5). “The clock on AI regulation is ticking”, warns UN rights chief. UN News. https://news.un.org/en/story/2026/10/1168529

United Nations. (2026g, September 25). Africa needs its own AI solutions, not borrowed ones, says expert. UN News. https://news.un.org/en/story/2026/09/1168440

United Nations High-level Advisory Body on Artificial Intelligence. (2024). Governing AI for humanity: Final report. United Nations. https://www.un.org/sites/un2.un.org/files/governing_ai_for_humanity_final_report_en.pdf