US and UK do not sign in Paris: how to tell what an international declaration requires
The US and UK absences matter politically, but the declaration is not a law. A method for reading signatories, verbs, scope and implementation.
On February 11, 2025, the Paris AI Action Summit ended with a declaration that the United States and the United Kingdom did not sign. The official list published by the Élysée names 63 countries—including China, India, France, Japan and Spain—and two organizations, the African Union Commission and the European Union. This was not “everyone except two”: delegations from more than 100 countries participated, but only those 65 signatories appear beneath the text.
The absence of Washington and London is politically conspicuous, although the document is not a treaty creating new legal obligations. Counting flags is not enough to understand its effect. Five elements matter: who signs, which verbs the text uses, what subject matter it covers, which institutions it creates and how performance will be checked. The same method works for any international communiqué, from AI to climate or trade.
First: open the document, not its headline
The declaration is titled “Statement on Inclusive and Sustainable Artificial Intelligence for People and the Planet.” Its official three-page English version lists six priorities: accessibility to reduce digital divides; open, inclusive, transparent, ethical, safe and trustworthy AI; innovation without excessive market concentration; positive effects on jobs and work; sustainability; and cooperation on international governance.
Its scope contains a limit often lost in coverage: a footnote says the statement concerns only civilian AI applications and uses. It does not govern military systems. It also contains no deadlines, sanctions, investment quotas or dispute mechanism. Phrases such as “we recognize,” “we underline” and “we look forward” dominate. These are political commitments and a shared agenda, not judicially enforceable commands.
That does not make the document useless. A declaration can align vocabulary, facilitate coalitions and announce infrastructure. Paris supports a public-interest AI platform and incubator, a future energy-impact observatory under the International Energy Agency and a network of labor observatories. But “launched” must be separated from “funded,” and welcoming an observatory from publishing data. Follow-through needs a budget, responsible bodies and dated outputs.
Second: a signature is not a law
International instruments sit on a spectrum. At one end is a binding treaty with ratification and defined duties. Other forms include decisions of an organization, political commitments, voluntary codes and statements of intent. The Paris document’s title and language place it in the political category. Signing expresses support; it does not automatically turn every priority into domestic law.
Who signs for whom also matters. The European Union appears beside member states, and the African Union Commission beside several African countries. This does not multiply obligations or mean every organization has identical powers. Readers should use the exact list rather than infer it from attendance. A leader’s presence, a group photograph or participation in a panel is not a signature.
Likewise, not signing does not prove rejection of every sentence. The British government did not publish an article-by-article reservation with the statement. Without such a primary source, reporting can establish its absence and attribute reasons supplied by a spokesperson, but should not reconstruct a complete doctrine. The United States did leave a broader public signal: in his full speech that day, JD Vance said he had come to discuss “opportunity,” not safety, advocated pro-growth policy and warned that excessive regulation could impede the industry.
Third: compare texts, not labels
Describing Paris as a total abandonment of safety also fails the document test. The text says trust and safety must advance and commends the Bletchley Park and Seoul summits. It promises further work on information integrity and transparency. The center of gravity changed: accessibility, employment, energy, market concentration and development were added to the frontier-risk agenda.
The November 2023 Bletchley Declaration, then signed by the United States, United Kingdom and China, devoted more attention to frontier models, hard-to-predict capabilities, cybersecurity and biotechnology risks, and the responsibility of developers of unusually powerful systems. It too was political and voluntary. Changing “Safety Summit” to “Action Summit” broadened priorities; it did not by itself alter the instrument’s legal nature.
A disciplined comparison records continuity and difference in a table. Object: Bletchley emphasizes advanced models; Paris covers the wider civilian ecosystem. Instruments: Bletchley discusses testing and evaluation; Paris adds platforms, observatories and capacity building. Implementation: both depend heavily on future cooperation. Signatories: they differ. This prevents a summit’s branding from becoming the conclusion.
Fourth: measure what happens afterward
Paris’s success will not be known from the signature count on February 11. It can be tracked through concrete questions. Is the public-interest platform created, and who governs it? Which projects does it fund? Does the energy observatory publish comparable methods for hardware, data centers and models? Does the labor network produce indicators distinguishing transformed tasks, created jobs and displacement? Do governments use those results in rules or procurement?
The US and UK absences must also be assessed through later conduct. A country may decline one text and still cooperate on safety testing, technical standards or research. Another may sign and contribute no money or data. A signature is a signal of alignment on one date; performance is a series of actions. Confusing them rewards declarative diplomacy and hides implementation.
For companies and citizens, immediate obligations still come from statutes, contracts and competent regulators, not this statement. A European company does not receive an exemption from the AI Act because the United States did not sign in Paris. Nor does the declaration create a new license for open models. It may guide future policy, but it cannot replace checking the applicable rule, effective date and enforcing authority.
A real division without exaggerating the document
The summit exposes a substantial political difference. The statement foregrounds coordination, sustainability and inclusion; Vance foregrounds leadership, infrastructure and deregulation. The United Kingdom remains outside despite having driven Bletchley. China signs both texts. This geometry shows that “safety versus innovation” does not split the world into two stable blocs: each government may support some formulations and reject others.
Paris leaves readers with a reusable protocol. Open the text; count signatories in the official list; locate scope and footnotes; classify the verbs; look for deadlines, money, institutions and oversight; then watch implementation. Those steps preserve the diplomatic importance of “not signing” without falsely turning it into repeal of a law, the end of all cooperation or an automatic victory for one power.
One final check concerns versioning. Official pages may update signatory lists or correct errors after a summit. Record the access date and, when available, preserve the dated PDF that was read. If two counts differ, do not choose the one that favors the story: explain which version produces each number. Documentary traceability is part of international policy too.
This article was produced with artificial intelligence under human editorial oversight.
