A Turkish fintech, SaaS company or bank runs on models it did not train, served from a cloud region abroad. This week the question “could Washington switch those models off?” left science fiction and entered founders’ group chats: after the 3 November 2026 midterms handed Democrats both chambers of Congress, a House member called for a pause on advanced AI and a ban on “artificial superintelligence” to be the party’s first order of business, and a bill with that content already sits in two committees. Without alarm or advocacy, this article explains what the bill says, how likely it is to move, what the rest of Washington is doing, whether a ban travels to the EU, China, the UN or Türkiye, and what a Turkish company built on frontier models should do.
What happened this week
On 3 November 2026 the United States held its midterm elections. Democrats won the House of Representatives and, by flipping seven seats, the Senate; as of 5 November they held at least 54 Senate seats with Alaska and Mississippi uncalled, and were projected to hold more than 250 of the 435 House seats. The new Congress is seated in January 2027.
On 4 November Representative Yassamin Ansari (Democrat, Arizona) posted on X that “top of the Democratic agenda on day 1 needs to be a pause on advanced AI development, regulation of this out of control industry, and a ban on artificial superintelligence”. Ansari is one of nine original cosponsors of the House version of the Ban Artificial Superintelligence Act of 2026, and her post set off the news cycle. She is one member of a majority of around 250, and neither the Democratic leadership nor any committee has adopted the bill as a programme: a marker, not a platform decision.
What the bill actually says
Senator Bernie Sanders (Independent, Vermont) and Representative Greg Casar (Democrat, Texas) announced the Act on 3 September 2026. S. 5493 was introduced on 23 September and referred to the Senate Commerce Committee; H.R. 10538 followed on 24 September, referred to the House Oversight and Government Reform Committee, with nine original cosponsors including Ro Khanna and Ansari (twelve by early November, Alexandria Ocasio-Cortez among the later additions). Neither bill has had a hearing or a vote. The introduced text, sixteen sections long, sets out this architecture.
| Element | What the bill would do |
|---|---|
| Scope | “Advanced artificial intelligence system”: any system trained with at least 1025 integer or floating-point operations; the Secretary adjusts the threshold annually (Section 3). |
| Regulator and pause | A cabinet-level Department of Artificial Intelligence. From enactment, training, modifying or fine-tuning advanced systems and deploying unreleased ones stops until the Department is fully staffed and has issued its safety rules (Sections 4 and 8). |
| Charters | Developing or distributing an advanced model requires a charter conditioned on full government access to systems, processes, staff and facilities; deploying, releasing, importing or transferring an advanced system needs a separate pre-deployment approval (Sections 9 and 12). |
| Permanent ban | No developing, deploying, acquiring, possessing, funding, importing or transferring “artificial superintelligence” (a system that exceeds, or can easily be modified to exceed, human cognitive performance across most domains or tasks, or that can plan and execute the destruction or disempowerment of humanity) or systems with “precursor characteristics” (automating AI research, resisting shutdown, self-modification, deceiving or evading oversight, uplift for nuclear, chemical or biological weapons, unauthorised access to secured infrastructure). Discovery is notified within 24 hours; precursor systems are paused and cut off from the internet at once and rendered inoperative within 30 days unless the characteristics are verifiably removed (Sections 9 to 11). |
| Sanctions | Up to 20 years’ imprisonment for covered policymaking individuals and “rogue actors” who recklessly violate the pause, prohibition or charter rules; a 10-year industry bar for other individuals; mandatory charter revocation, with IP receivership pending appeal and forfeiture if upheld (the sponsors’ “corporate death penalty”); whistleblower protection; no federal funds for violating activity (Sections 13, 14 and 16). |
| Outward policy | A declared US policy to prevent superintelligence worldwide through international agreements and allied coordination, which “may include export controls on artificial intelligence computing infrastructure” (Section 15). |
How likely is it to move?
The bill was introduced under a Republican Congress and did not move. The new majority controls the committee gavels, so a hearing in 2027 is plausible. Passage is another matter: a bill halting frontier training in the United States while competitors continue would need the 60 Senate votes most legislation still requires, then survive a presidential veto, which takes two-thirds of both chambers to override. Even within the Democratic caucus, press reports note, support for a bill carrying 20-year prison terms is not guaranteed. In our view the odds moved from near zero to low, and whatever reaches the floor in 2027 may share a title with S. 5493 and little else: a signal about one wing of the new majority, not a timetable.
The opposite current in the same city
The executive branch has spent 2026 building the alternative model. Executive Order 14409 of 2 June 2026, “Promoting Advanced Artificial Intelligence Innovation and Security” (91 FR 34565), sets up a classified benchmark to designate “covered frontier models”, invites developers to give the government access for up to 30 days before release to other trusted partners, and states that nothing in it authorises mandatory licensing, pre-clearance or permitting. On 29 September 2026 Executive Order 14434 told agencies to say “super intelligence” (SI) instead of AI and gave the President’s science adviser 60 days to propose a legal definition; the same day the chief executives of Google, Anthropic, Meta, xAI and NVIDIA and the president of OpenAI signed a voluntary White House accord (internal capability controls, external auditors, an independent review committee). Vice President Vance rejected an FDA- or FAA-style regulator: the FTC and the Department of Justice, he said, already have the authority they need.
Congress blocked the middle ground: in September single objections stopped a Republican “AI Emergency Button Act” (a company-controlled kill switch) and a Democratic bill for an AI Safety Board with 45 days of pre-release model access, while Florida’s Attorney General asked a state court on 28 September to bar OpenAI from building new models without independent safety guardrails. During Xi Jinping’s September state visit the United States and China agreed a “Super Intelligence Dialogue” and an incident channel, even as the White House science adviser rejected any “globalist scheme of control”. Washington is not of one mind.
Public sentiment is, in our view, the ban’s main asset. The Future of Life Institute’s Statement on Superintelligence, launched in October 2025 and signed by Geoffrey Hinton, Yoshua Bengio, Steve Wozniak, Steve Bannon and Prince Harry among others, calls for “a prohibition on the development of superintelligence, not lifted before there is broad scientific consensus that it will be done safely and controllably, and strong public buy-in”; by early November 2026 it had more than 78,000 signatures. The Institute’s own US poll puts support for holding off until scientists agree it is safe at 64 per cent; it is an advocacy group’s poll.
Does the idea of a ban travel?
Compare the jurisdictions a Turkish company actually meets.
European Union. The EU uses the same 1025 number to presume systemic risk in a general-purpose model (Article 51(2)) and chose obligations rather than prohibition: the metric has converged, the consequence has not. Chapter V of the AI Act has bound providers of general-purpose models since 2 August 2025; since 2 August 2026 the Commission’s AI Office can demand documents, run evaluations, order corrective measures including market withdrawal, and fine providers up to 3 per cent of worldwide turnover or EUR 15 million, whichever is higher (Articles 88 to 93 and 101). The Digital Omnibus postponed the high-risk timetable but not these GPAI rules or the Article 50 transparency duties. In her State of the Union speech on 16 September 2026, Commission President von der Leyen said that “CEOs of the most advanced companies tell us that it is time to slow down on the self-recursive models. To pace the frontier”, and promised to “team up on model evaluation, verification, early warning, AI security” with Canada, the United Kingdom and like-minded partners. No pause or ban was proposed.
China and the UN. China has domestic generative-AI rules and, since 1 September 2025, requires AI-generated content to be labelled; whether Beijing would slow its own labs is contested, and analysts who follow Chinese compute generally argue it will not. The UN Global Dialogue on AI Governance first met in Geneva on 6 and 7 July 2026, informed by the Independent International Scientific Panel’s preliminary report of 1 July 2026; it is expressly “not a negotiating forum”, and its second session is set for 3 and 4 May 2027 in New York. The non-proliferation treaty analogy used by the bill’s sponsors has no draft text anywhere.
Türkiye. No ban is on the table. The draft AI law (2/2234), before Parliament since June 2024, is risk-based; the AI Action Plan 2026-2030 (Presidential Circular 2026/9, Official Gazette 18 August 2026, No. 33344) is a pro-adoption programme that, as reported, foresees regulatory sandboxes and tiered oversight; and Article 11(1)(g) of the Personal Data Protection Law No. 6698 already lets individuals object to a result produced against them exclusively through automated analysis. Türkiye has no 1025 training runs to pause: a US pause or export control would reach Turkish companies through their suppliers’ terms and cloud regions, not through Turkish law.
In our view the ban travels as vocabulary and supply-chain pressure, not as law. Exporting it is hard for structural reasons: “superintelligence” has no agreed definition; compute thresholds age so fast that the bill itself provides for annual adjustment; enforcement presupposes domestic frontier labs; and US-China competition cuts against unilateral restraint. Where it can travel is narrower but real: export controls on chips and cloud capacity; providers’ acceptable-use terms and shutdown rights, which bind every downstream customer by contract; government incident channels; and political vocabulary, as the “super intelligence” rename shows.
Where to start
- Read the supply contract for pause and withdrawal scenarios. What happens if your provider must suspend, geofence or withdraw a model under a US order, an EU corrective measure or its own safety policy: notice, continuity, migration assistance, data export. See our guide to negotiating an AI vendor contract.
- Design for more than one provider. An abstraction layer, portable prompts and evaluation suites, and a tested fallback model turn a legal risk into an engineering task.
- Comply with the rules that bind today. If you serve EU users, the AI Act reaches you from Türkiye and building on a foundation model does not remove your own GPAI-related duties; Article 50 transparency has applied since 2 August 2026, with a marking grace period to 2 December 2026 for systems placed on the market before that date.
- Keep the Turkish governance layer in order. Board approval and documentation of AI adoption under the Commercial Code duty of care, a process for objections to automated decisions under KVKK Article 11, and a system inventory with model cards.
- Keep a 2027 watch list. The new Congress’s first AI bills and any hearing on S. 5493 or H.R. 10538; the “super intelligence” definition due by late November 2026; the AI Office’s first enforcement actions; Türkiye’s draft law; the US-China dialogue.
Frequently asked questions
Has the United States banned superintelligence?
No. The Ban Artificial Superintelligence Act of 2026 (S. 5493 and H.R. 10538) was introduced in September 2026 and sits in committee without a hearing or vote. Executive policy, set by Executive Order 14409 and the voluntary accord of 29 September 2026, expressly rejects mandatory licensing. One member’s call to prioritise the bill does not change the law.
Could a US pause reach a Turkish company?
Not directly: a US statute cannot regulate a company in Türkiye. It could reach Turkish users indirectly if a provider had to suspend or withdraw a model, if export controls restricted access to chips or cloud capacity, or if provider terms changed. The practical protection is contractual (notice, continuity and migration clauses) and architectural (more than one provider).
Is Türkiye considering a similar ban?
No such proposal exists. The pending draft AI law is risk-based, the 2026-2030 AI Action Plan promotes adoption, and the Personal Data Protection Law already governs automated decisions about individuals. What matters for Turkish companies is the EU rules binding their suppliers and customers, and the contractual terms of the models they use.
Sources. S. 5493 and H.R. 10538, Ban Artificial Superintelligence Act of 2026, 119th Congress, introduced text (congress.gov, govinfo.gov); Sanders press release, 3 September 2026; PBS NewsHour / AP, 23 September 2026; Executive Order 14409 of 2 June 2026, 91 FR 34565 (Federal Register); Executive Order 14434 of 29 September 2026, “Inaugurating the Era of Super Intelligence” (Federal Register) and White House accord (Nextgov); Sen. Kennedy release, 16 September 2026; WLRN on the Florida filing, 29 September 2026; TNW on the US-China SI Dialogue, 28 September 2026; CBS News Senate results, 5 November 2026 and NBC midterm results, 4 November 2026; The Standard on Rep. Ansari’s post, 5 November 2026; Statement on Superintelligence and Future of Life Institute press release; Regulation (EU) 2024/1689 (AI Act), Articles 50, 51, 88 to 93, 101 and 113 (EUR-Lex); Euronews on the State of the Union speech, 16 September 2026; UN Global Dialogue on AI Governance and Scientific Panel preliminary report, 1 July 2026; Law No. 6698 on the Protection of Personal Data, Art. 11 (mevzuat.gov.tr); Presidential Circular 2026/9, Official Gazette 18 August 2026, No. 33344; Draft Law 2/2234, Grand National Assembly of Türkiye.
This article is general information and not legal advice; a specific situation should be assessed with counsel.
