In recent days, this call has been voiced by those who lead those laboratories. In the essay “We Must Pace the Frontier”, published in September, Anthropic co-founder and CEO Dario Amodei wrote that “we must slow the pace at which we improve the capabilities of AI models”, citing as causes the acceleration of recursive self-improvement in AI systems and recent episodes of agentic misalignment, and estimating at six to twelve months the window within which a
sufficiently capable swarm of agents could come to compromise significant parts of the network. The proposal unfolds in three steps: third-party evaluators embedded in the laboratories with continuous access; coordination among democracies on common safety standards; and global coordination, up to a shared “speed limit” on recursive self-improvement and verifiable agreements with the People's Republic of China. “I believe we owe it to humanity to try”, Amodei concludes. Two of the three steps are multilateral by definition, and today they have no institutional home in which to be negotiated, verified and legitimized. That is precisely the home Milan offers to build.

On 2 September 2026, in Chapel Hill, the G20 Innovation Ministers adopted by consensus the Carolina Principles for Emerging Technologies, also endorsed by the People's Republic of China. The framework promotes a sectoral, proportionate and adaptable approach, geared towards human-centric innovation and the safe and trustworthy adoption of AI, while avoiding the creation of new general-purpose bureaucracies. The heads of the leading frontier laboratories took part at the same table, including Anthropic co-founder Tom Brown. The G20 framework completes an architecture on the side of States' industrial and regulatory policy; by an explicit choice of scope, it does not cover the properly multilateral level or the dimension of ethical principles and fundamental rights, which remain within the remit of the United Nations. It is precisely in that complementary space that the Milan pillar is positioned.

On 29 September 2026, at the White House, US President Donald Trump and the heads of Google, Meta, Anthropic, OpenAI, Nvidia and xAI signed the White House Accord on Super Intelligence, a “Joint Commitment on Frontier Responsibilities”. The Accord, voluntary and not legally binding, commits each company training and deploying frontier models to four layers of controls and audits: internal controls on the capabilities and alignment of its models, with particular attention to cybersecurity, biosecurity and chemical threats; an internal team to ensure that those controls operate as intended; independent assessments by an external auditor or evaluator; and an independent committee of the board of directors to oversee the results. The text acknowledges that over time it may make sense to codify these steps into laws or regulations. It is the laboratories that build the frontier, together with the government of the country that hosts them, that now recognize independent evaluation as the condition of trust: the first of the three steps set out by Amodei thus becomes part of the companies' own commitments. By its nature, the Accord remains a national and voluntary commitment among companies, and it leaves open the question of how to make the methods of that evaluation comparable across countries. It is to this question that the functions of the Milan pillar respond, measuring without regulating, in full respect of the sovereignty of States.

Milan has the credentials to occupy that space. Assolombarda forecasts growth of 1.7% for the city in 2026, more than 8 points above pre-Covid levels. In 2024 the area attracted €2.3 billion in venture funding. Seven Milan data centres have been recognized as of national strategic interest. Bending Spoons listed on Nasdaq at over $25 billion. Politecnico, Statale, Bicocca, Bocconi, Human Technopole, Mario Negri and CNR form a research ecosystem of European critical mass. The Vatican, one hour away by plane, has already offered, with the Note Antiqua et nova of January 2025, a moral anchor that no other AI capital can claim.

Milan and Lombardy are also an example of a research ecosystem combined with a
manufacturing heartland. This is a central point, because the phase now opening no longer concerns software alone. Physical AI, the artificial intelligence that perceives its environment, decides and acts in the real world through robots, autonomous machines, and intelligent systems and infrastructure, will profoundly affect people, societies and businesses. Milan is a
natural place to test this transition to intelligence that acts in the real world.

Alongside this is the city's ongoing experience as an urban laboratory for responsible AI. The Municipality of Milan has promoted a Manifesto for the use of artificial intelligence, a pioneering experience among European cities, consistent with the principles of the AI Act, and has mapped 45 AI use cases already in production within the Administration and in city services. The Smart City Lab and the Human AI+ programme, now in its second edition, translate those principles into pilots and businesses, while the Board for Technological Innovation and Digital Transformation brings together the public and private sectors, universities and the third sector in an open governance method. Milan does not offer an empty seat. It offers an ecosystem in which the principles of responsible AI are already being tested in the complexity of a major European city.

The 2017 EMA precedent teaches us that the best of technical dossiers, with Milan first in the first round of voting and victorious in the second, can be lost in the final stretch if the country is not united and bilateral relations with the decisive Member States do not start in time. We will not repeat that mistake.

There is a coincidence that cannot be overlooked. The architecture of multilateralism was born in San Francisco on 26 June 1945 with the signing of the Charter of the United Nations. Eighty years later, San Francisco is the world capital of the industrial revolution of artificial intelligence. The same place that saw the birth of the UN now sees the birth of the technology that, more than any other, challenges the international order of the twentieth century. Both respond to the same question: how to govern a discontinuity that no State can face alone.

This is where the Sputnik Moment of the United Nations begins. At a time of profound rationalization of the Organization, the governance of artificial intelligence is not one more agency to be added to the forty-five already in existence. It is the horizontal vector around which the UN can regain its historic centrality: the new-generation pillar, distinct from the twentieth-century agency model, that restores meaning to the universalism of the United Nations in the twenty-first century. It is within this logic of rationalization, consistent with the UN80 approach and with the light-touch spirit of the Carolina Principles, that the pillar neither duplicates existing agencies nor introduces new regulatory burdens, but connects them and provides the horizontal layer that the system currently lacks.

The pillar's line of responsibility runs along the United Nations' own remit. Operational support to the Global Dialogue on AI Governance and to the Independent Scientific Panel established by Resolution A/RES/79/325. Interoperability among States' regulatory frameworks, in continuity with the sectoral and adaptable approach recognized by the G20. Capacity-building for the Global South, so that the rules that concern everyone are also written by those who do not sit at the table today. Independent evaluation of public-sector AI pilots and of their impact on work and services. Stewardship of the ethical dimension and of fundamental rights, rooted in the Magisterium of the Catholic Church and in the European constitutional tradition, which no other international forum explicitly covers.

The twelve functions of the pillar

AmCham Italy's Policy Paper II, “Milan and the future of global governance of artificial intelligence. The functions of the new UN pillar for AI governance”, translates this line of responsibility into twelve technical functions. They are not an invention: they make operational the principles of the UNESCO Recommendation on the Ethics of Artificial Intelligence, adopted in 2021 by all Member States, within the framework of Resolution A/RES/79/325, and they are ordered by a single criterion: technology must remain at the service of the person. Each has a declared limit and a performance indicator, and is exercised together with the organization competent for the subject matter, which retains ownership of it.

1. Pace of the frontier and early warning

   Measure with a public method how fast the capabilities of the most advanced systems are growing, and alert States in time to those with systemic implications.

2. Impact assessment

   A common method, updated over time, to assess the effects of a system on people.

3. Human responsibility

   Whoever answers for a decision always remains identifiable along the technological chain.

4. Transparency and provenance

   Distinguish the authentic from the synthetic and obtain explanations proportionate to the decision.

5. Data and documentation

   Document which data a system was trained on and on what basis they were acquired.

6. Equity among countries

   A comparable index of gaps in compute, data and skills, in order to reduce them.

7. Labour and transition

   Comparable data on employment effects and on the transition policies that work.

8. Health and minors

   Comparable evidence and precautionary criteria for the most vulnerable people.

9. Training regulators

   Skills for those who regulate and those who use AI, the precondition for all other functions.

10. Languages and pluralism

    Measure which languages and cultures systems represent, and support those left out.

11. Environment and compute

    Make the energy, water and land footprint of data centres comparable.

12. Security and stability

    A common language to recognize and report incidents and dangerous capabilities.

Four products: methods, measures, warnings, skills · No normative, authorizing or sanctioning power

None of the four products is a rule and none entails compliance obligations. The pillar does not authorize or register models, does not supervise or sanction companies, does not produce horizontal rules, and takes no competence away from ILO, WHO, UNESCO, UNDP, UNCTAD, ITU, WIPO and OHCHR, nor the scientific role from the Independent Panel. Implementation proceeds in three phases: within twelve months, the functions that require no new mandates; from twelve to thirty-six months, consolidation; beyond thirty-six months, the functions that require scale. It is the space where those who ask for less regulation and those who ask for more verification can meet: measuring does not mean prohibiting; it means making every subsequent choice possible.

The positioning of Milan's candidacy is today clearly the leading one in the West. Since last March, the Republic of Korea has been pursuing the physical consolidation into a single campus of the AI functions of six existing UN agencies, a twentieth-century agency model. On 16 July, China launched in Shanghai the World AI Cooperation Organization, an alternative intergovernmental organization signed by twenty-nine States and placed outside the UN framework. Kazakhstan has offered Astana as the venue for that body's first meeting. Milan is the first European city to put itself forward as the seat of a new-generation pillar within the United Nations framework, and the first in the world whose candidacy originates from a civic and business actor rather than from a government. It is a positioning without precedent.

The governance of artificial intelligence cannot be reduced to a table reserved for the technologically dominant economies alone, with Europe and the Global South relegated to spectators of the rules that concern them. This is not in opposition to the United States and China, essential and complementary interlocutors in any credible multilateral framework, but because the horizontal and universal nature of the technology requires an inclusive table. Milan is putting itself forward as the European seat of that table, where the method of law, fundamental rights and institutional pluralism, already codified by the AI Act, is added as a
third constitutive axis of the global governance of artificial intelligence. It is the space in which Europe can once again compete, build and grow on the terrain that, more than any other, will define the continent's competitiveness in the coming decades, restoring to the Union an institutional centrality that the twentieth-century agency model had not granted it.

The most innovative feature of the proposal concerns its financial structure. The pillar would rest on a three-tier hybrid model, comprising assessed contributions from Member States, permanent voluntary contributions from the sector's leading operators, on the successful model of GAVI and CEPI in global health, and thematic partnerships with philanthropic foundations and international financial institutions. The model reduces pressure on the Organization's regular budget and offers a useful institutional precedent for the financial reform of the UN system.

For these reasons, the undersigned commit to

Publicly and institutionally support

Milan's candidacy as the seat of the new United Nations pillar for the governance of artificial intelligence, set out in the policy paper “Milan, global capital of ethical artificial intelligence” presented by AmCham Italy on 27 July 2026 and developed into the twelve functions of Policy Paper II of September 2026.

Establishing a standing Promoting Committee

with its operational headquarters in Milan, bringing together the business, commercial, academic, scientific, professional and financial representatives of the Milan and Lombardy area, and engaging in ongoing dialogue with the Presidency of the Council of Ministers, the Ministry of Foreign Affairs (Farnesina), the Ministry of Enterprises and Made in Italy (MIMIT), the Ministry of Justice, the Lombardy Region and the Municipality of Milan.

Contribute their expertise

to the technical dossier of the candidacy, each according to its own institutional mission, providing research, data, sector experience and the capacity to engage with international stakeholders, and making available to the dossier the initiatives through which the Milan and Lombardy production system is already testing and industrializing artificial intelligence applied to the physical world.

Support the diplomatic action

of the Italian Government with the United Nations Secretariat, the Permanent Mission of Italy to the United Nations in New York and the decisive Member States, from the co-facilitators Spain and Costa Rica to the leading members of GRULAC, the African Group and the G77, in view of the high-level week of 22–28 September 2026 and the May 2027 checkpoint.

Ensure consistency with the G20 architecture

in dialogue with the major frontier laboratories and the competent international organizations, so that the Milan pillar is a lean structure, complementary to the Carolina Principles of 2 September 2026 and to the voluntary commitments of the White House Accord on Super Intelligence of 29 September 2026, avoiding any duplication and serving as the UN seat for the ethical, human-centric and independent technical-capacity dimension that the G20 framework explicitly does not cover.

Recognize the urgency of the time window

The window for positioning Milan is the next twenty-four months. Those who come to that table with a ready seat, a credible technical dossier, a sustainable financial structure and a cohesive political coalition will shape the institutional geography of AI governance for the next generation. Those who arrive later will negotiate on terms written by others.

The meaning of this signature

We are not signing a petition. We are signing an operational commitment, with the same seriousness with which Diana Bracco's coalition supported Milan's EMA candidacy in 2017, and with the added awareness that what is at stake today is the governance of the technology that will redefine the economy, work, information and democracy over the next twenty years.

Every signatory carries equal weight. The Manifesto is promoted by AmCham Italy, Assolombarda and Confcommercio Milano Lodi Monza Brianza, and the Promoting Committee is civic, cross-cutting and non-partisan. No logo prevails over the others; no institution speaks for the others. The dossier belongs to Milan, not to a single house.

Milan is not asking for a privilege. It offers the country a concrete opportunity to play a leading role in building the international architecture that will govern artificial intelligence. It also offers a place where AI governance can maintain a direct relationship with its application. Effective governance will help create the conditions for AI to be used safely, responsibly and positively to generate innovation, well-being and growth. The proposal is on the table. Let it be discussed at the appropriate levels, with the urgency and seriousness the moment demands.

Alea iacta est.