On 22 September 2026, in his address to the UN General Assembly, President Trump said the United States "totally rejects any attempt to construct a globalist scheme of control for the Artificial Intelligence being spoken of so much now — hereinafter officially called 'Super Intelligence.'" He added that he would "only encourage super intelligence… not rein it in."
Nothing in that speech changed a single compliance deadline. The EU AI Act's staged dates, Japan's AI Act and guidelines, and the US state laws that actually bind companies were all identical the day after the speech. Here is the short version of where the three jurisdictions stand: the US has no comprehensive federal AI statute (executive orders plus state law), the EU has one binding regulation with fines up to €35 million or 7% of global turnover, and Japan has a promotion act with no penalty provisions at all.
This article compares the three regimes using primary sources only — the White House release, the US Federal Register, EUR-Lex, Japan's Cabinet Office and ministries, UN General Assembly resolution A/RES/79/325, and OECD publications — as verified on 23 September 2026. It makes no political assessment. It is not legal advice; confirm your own position with counsel.
What Trump actually said about AI regulation
Two facts are verifiable from the official record. First, the US stated it rejects participation in an international control framework for AI. Second, the US said its domestic posture is to encourage, not restrain, the technology. The renaming itself applies to "all United States documents" — federal government documents — not to private companies, other governments or academic usage.
The rest of the announcement, including the Truth Social polls that produced the name, is covered in our piece on Trump renaming AI to "Super Intelligence" and what it means. If you are unsure how the term relates to AGI and ASI as researchers use them, see the difference between AGI, ASI and SI in one table.
As of 23 September 2026, we could not find an executive order or Federal Register notice implementing the rename. It remains a statement made in a speech.
US vs EU vs Japan: AI regulation at a glance
Two questions decide how much any of this matters to you: is it enforceable, and does it reach you. Sorted that way, the three regimes look like this.
Dimension | United States | European Union | Japan |
|---|---|---|---|
Core instrument | Executive orders plus state laws and city rules; no comprehensive federal statute | EU AI Act, Regulation (EU) 2024/1689, amended by 2026/1744 | AI Act (Act No. 53 of 2025) plus non-binding principles and business guidelines |
Binding on companies? | Yes at state level; federal orders mainly direct agencies | Yes — directly applicable in all member states | No penalty provisions in the statute |
Penalties | Varies by state (California SB 53: up to $1M per violation) | Art. 99: up to €35M or 7% of worldwide turnover, whichever is higher | None under the AI Act |
Regulatory design | States legislate, the federal level pushes back | Four risk tiers by intended use, obligations scaled per tier | Promotion-first; real risk sits in existing laws (privacy, copyright) |
Extraterritorial reach | Many statutes reach anyone serving that state's residents | Art. 2(1): applies where AI output is used in the EU, wherever you sit | No extraterritorial mechanism |
Nearest real deadline | 1 Jan 2027 — California ADMT compliance begins | 2 Dec 2026 — marking of synthetic content for pre-existing systems | Amended privacy law in force on a date yet to be set (enacted 17 Jul 2026) |
If you read one row, read the penalties row. Only the EU and a handful of US states attach money to non-compliance today.
United States: executive orders that direct agencies, state laws that bind you
The "don't regulate, encourage" posture predates the UN speech and already exists in writing.
- Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence" (signed 11 December 2025, published 16 December 2025). It directs the Attorney General to stand up a task force to challenge state AI laws that conflict with federal policy.
- Executive Order 14409, "Promoting Advanced Artificial Intelligence Innovation and Security" (signed 2 June 2026, published 5 June 2026). Section 1 states the US leads in AI "because we refuse to stifle this innovation with overly burdensome regulation." Its operative sections direct agencies on cyber defense of national security systems.
The distinction that matters in practice: an executive order instructs the executive branch. It does not by itself void a state law or create duties for private companies. Overriding state law requires a court ruling or federal legislation, and no preemptive federal AI statute has been enacted.
What does bind companies is state law: California AB 2013 (1 Jan 2026), Texas HB 149 (1 Jan 2026), Illinois HB 3773 (1 Jan 2026), and New York City Local Law 144 (enforced since 5 July 2023). Most of these reach you based on who you serve, not where you are incorporated.
EU AI Act: the only binding comprehensive regime — with later deadlines than you may think
The AI Act entered into force on 1 August 2024 with staged application under Article 113. Regulation (EU) 2026/1744, the Digital Omnibus on AI adopted 8 July 2026, moved several of those dates. The current picture:
- 2 February 2025 — prohibited AI practices (Art. 5) and AI literacy (Art. 4) already apply.
- 2 August 2025 — general-purpose AI obligations, governance and the penalties chapter already apply.
- 2 August 2026 — general application, including Article 50 transparency duties.
- 2 December 2026 — deadline for machine-readable marking of synthetic content by providers whose systems were already on the market.
- 2 December 2027 — Annex III high-risk obligations (hiring, credit, education and more), postponed from 2 August 2026.
- 2 August 2028 — high-risk AI embedded in Annex I regulated products.
A detail that repeatedly gets lost: Article 99(6) caps fines for SMEs, including startups, at the lower of the fixed amount and the turnover percentage. The headline €35M figure is not what a small company faces.
Japan: a promotion act with no penalties, and the rules that actually bite
Japan's AI Act (Act on the Promotion of Research and Development and Utilization of AI-Related Technologies, Act No. 53 of 2025) was promulgated on 4 June 2025 and came fully into force on 1 September 2025. Article 7, the provision aimed at businesses, asks companies to use AI actively and to cooperate with national policy. There are no penalty or administrative sanction provisions in the statute.
The operative substance sits in soft law: the government's principles on ensuring appropriateness in AI R&D and use (adopted 19 December 2025 under Article 13) and the AI Guidelines for Business version 1.2 (MIC and METI, 31 March 2026). The principles explicitly state that the expected level of response scales with each organisation's size, role and risk — so a small company is not expected to build enterprise-grade governance.
Legal exposure in Japan comes from the general laws. The amended Act on the Protection of Personal Information was promulgated on 17 July 2026, adding administrative surcharge orders. Its commencement date is "within two years of promulgation, on a date set by cabinet order," and that order had not been published as of 23 September 2026.
What the UN framework actually does — and does not — require
This is the part most commentary gets wrong. Read against the text, the UN machinery does not impose obligations on anyone.
UN General Assembly resolution A/RES/79/325, adopted 26 August 2025, created two bodies:
- The Independent International Scientific Panel on AI — 40 members serving in their personal capacity for three-year terms. Its mandate is to issue "one annual policy-relevant but non-prescriptive summary report" plus thematic briefs. The 40 members were appointed on 12 February 2026 and elected co-chairs at the inaugural meeting on 3 March 2026.
- The Global Dialogue on AI Governance — convened annually for up to two days, alternating between Geneva and New York. Its output is a Co-Chairs' summary. The first Dialogue took place in Geneva on 6–7 July 2026; the second is scheduled for New York in 2027.
The resolution also limits the Panel and Dialogue to the non-military domain. In other words, what this framework produces is reports and summaries. As of 23 September 2026 there is no UN instrument that binds states or companies on civilian AI.
G7 Hiroshima AI Process: voluntary reporting, now aimed at SMEs
The G7 track has the same character. The Hiroshima Process International Code of Conduct, launched under Japan's 2023 G7 presidency, is monitored through an OECD reporting framework that is explicitly voluntary. On 28 May 2026 the OECD launched version 2.0 at a Tech7 event in Paris under the French G7 presidency, streamlining it so that deployers and small and medium enterprises — not just large model developers — can participate. More than 50 companies have pledged to complete it.
In practice, this is the cheapest way for a mid-sized company to demonstrate AI governance to enterprise customers without an EU-style conformity assessment.
Does the speech change what you have to do? No — here is what does
Work backwards from your own situation and the list of regimes you need to read shrinks fast.
Your situation | What applies | Date to watch | First move |
|---|---|---|---|
Domestic market only, using off-the-shelf AI internally | Privacy and copyright law; national AI guidelines | Your local privacy amendment | One page: approved tools, and data that must never be pasted in |
Shipping a product with generative AI that talks to users | EU AI Act Art. 50; US state disclosure rules | 2 December 2026 | Put AI disclosure into the UI and the terms of service |
Using AI in hiring, credit, pricing or termination decisions | EU Annex III; Colorado, California, Illinois, NYC | Already in force in NYC (2023) and Illinois (1 Jan 2026); 1 Jan 2027 (Colorado, California); 2 Dec 2027 (EU) | Map where AI touches adverse decisions and build a human-review path |
Training your own foundation models | California SB 53; EU GPAI duties; HAIP reporting | Both already in effect | Check the compute and revenue thresholds before anything else |
A common pattern we see: companies in row three assume they are too small to be in scope. They are not. Hiring, credit and termination are the first use cases every jurisdiction regulates, and the EU, several US states and New York City all landed on them independently.
One more practical note. You do not need to rewrite contracts, policies or IR documents to say "SI." Statutory terminology has not changed anywhere, and a rename announced in a speech is not a drafting instruction. We work through that question in detail in whether to update contracts and internal policies after the AI-to-SI rename, and define the term itself in what "Super Intelligence" (SI) actually means.
What we could not confirm
- Any executive order or Federal Register notice implementing the rename (not found as of 23 September 2026).
- Official responses from the European Commission or the Japanese government to the speech.
- Whether the US intends to withdraw from the UN Scientific Panel or Global Dialogue; no change appears in UN materials.
- The cabinet order setting the commencement date of Japan's amended privacy law.
We have left these open rather than guessing, and will update the article when they are published.
At Mihata we help companies inventory the AI already in use across their teams and design where AI belongs in a workflow. Cross-border legal analysis belongs with your counsel, but the underlying question — which tool touches which decision, and on what basis — is answered fastest by the people who run the process. We are happy to start from wherever you are.
FAQ
Did Trump's UN speech change any AI compliance obligation?
No. The speech announced that US government documents will use the term Super Intelligence and stated that the US rejects an international control framework for AI. It did not amend the EU AI Act, Japan's AI Act or any US state law. As of 23 September 2026 no executive order or Federal Register notice implementing the rename could be found.
What is the international AI framework that the US rejected?
UN General Assembly resolution A/RES/79/325, adopted 26 August 2025, created the Independent International Scientific Panel on AI and the Global Dialogue on AI Governance. Under the text the Panel issues one annual policy-relevant but non-prescriptive summary report, and the Dialogue produces a Co-Chairs' summary. Neither imposes obligations on states or companies.
Which jurisdiction regulates AI most strictly: the US, the EU or Japan?
The EU, measured by binding force and penalties. The AI Act is directly applicable across member states and Article 99 sets fines of up to 35 million euros or 7 percent of worldwide turnover, whichever is higher. The US has no comprehensive federal statute and relies on state law, while Japan's AI Act contains no penalty provisions.
Does the EU AI Act apply to a company outside the EU?
It can. Article 2(1) covers providers placing AI systems on the EU market regardless of where they are established, and also covers providers and deployers in third countries where the output of the AI system is used in the EU. Being headquartered elsewhere is not, on its own, a reason to be out of scope.
Is the G7 Hiroshima AI Process mandatory?
No. The OECD reporting framework that monitors the Hiroshima Process International Code of Conduct is voluntary. Version 2.0 launched on 28 May 2026 and was streamlined so that deployers and small and medium enterprises can participate, with more than 50 companies pledging to complete it.