The world's governments are racing to establish shared rules for artificial intelligence before the technology outpaces their ability to govern it. The first session of the UN Global Dialogue on AI Governance represents the only universal forum where democratic deliberation on AI's future can happen at the international level, according to a submission from the Centre for Future Generations (CFG) to the dialogue. The stakes are enormous: decisions made in the next few years about how AI systems gain autonomy, how they're deployed, and who has access to them will shape the conditions under which billions of people live and work for decades to come.

The core problem, as CFG frames it, operates on two fronts. First, there's the tension between technology companies, which shape finance and markets, and nation states trying to catch up with enforcement. Second, there's AI as an instrument of geopolitical competition, particularly between the United States and China, centered on control of global data and advanced computing resources. Both tensions share a common diagnosis: the ungoverned transfer of autonomy from humans to AI systems whose behavior remains opaque, unaccountable, and difficult to reverse.

What Would Make the First Global AI Dialogue a Success?

CFG proposes that a successful first dialogue should produce three concrete outcomes: a shared problem statement, a working methodology, and a follow-up architecture. The organization suggests that the dialogue's seven thematic areas should cohere around a single question: at what level of autonomy is each AI system operating relative to its users, and is that level governed?

To measure this, CFG proposes "autonomy-gap analysis," a methodology that measures the gap between the autonomy a law permits and the autonomy a system actually exercises in deployment. The organization has already built an Enforcement Tracker showing this gap for the EU AI Act, the Digital Services Act, and the Digital Fairness Act. If adopted at the UN level, this methodology would let governments report on enforcement reality rather than just legislative intent.

How to Build International AI Governance Frameworks

• Establish Common Audit Methods: Create an intergovernmental expert group to develop shared audit methodologies across cognitive, security, and labor domains where AI autonomy transfer poses the greatest risks.
• Create Reporting Lines to Existing Bodies: Connect the dialogue's findings to established UN agencies including the Office of the High Commissioner for Human Rights (OHCHR), the International Labour Organization (ILO), and UNESCO, which already have mandates covering human rights, labor, and education.
• Build a Public Registry: Establish a public registry of national autonomy-gap findings so governments can transparently report how well their AI regulations are actually working in practice.
• Protect Cognitive Integrity: Adopt "cognitive integrity" as a shared governance standard, defined as the condition where people think, decide, and act based on their own goals and values rather than being manipulated by external AI systems.

What Critical Issues Are Missing From Current AI Governance Discussions?

CFG identifies five cross-cutting issues that current governance frameworks overlook. The first is intergenerational fairness: today's AI governance choices impose irreversible commitments on people who don't yet have a vote. Autonomy transfer compounds over time, as systems permitted to operate at a given level today shape the workflows and institutional practices that constrain what's reversible tomorrow.

The second issue is compute concentration. Taiwan Semiconductor Manufacturing Company (TSMC) fabricates most advanced-node logic that AI runs on, ASML is the sole supplier of extreme ultraviolet lithography equipment, and NVIDIA controls most of the AI accelerator market. Governance of safety and human rights at the application layer cannot bind decisions made at the substrate layer by firms answerable to neither the user state nor the host state.

The third is the civil-military boundary. The fastest autonomy transfer is happening in defense and security applications, where the EU AI Act and most national frameworks don't reach. NATO, the Organization for Security and Co-operation in Europe (OSCE), and the UN First Committee have parallel conversations the dialogue doesn't currently connect. CFG recommends adding a reporting line into the Office for Disarmament Affairs to close the gap between civilian governance and the domain where ungoverned autonomy carries the highest immediate risk.

The fourth issue is planetary boundaries. Current governance frameworks leave AI's energy and water consumption exogenous to the discussion, despite the massive resource demands of training and running large AI systems. The fifth is cultural convergence: systems trained on similar datasets and designed by similar people narrow what gets reproduced at scale, making cultural diversity a governance question.

How Is AI Being Integrated Into Medicine Regulation?

While international governance frameworks are still being negotiated, individual sectors are moving ahead with AI integration. The European medicines regulatory system is transitioning to a more data-driven ecosystem where artificial intelligence and advanced analytics help regulators interrogate and reuse large amounts of data from multiple sources, including clinical trials, real-world healthcare databases, patient registries, and digital health platforms.

The European medicines regulatory network (EMRN) has established a governance structure and strategies for data standardization to build what it calls a "trusted ecosystem." Over the last five years, the network established the Big Data Steering Group in 2020 to promote better analysis and use of big data in medicines regulation. This work has now expanded under the newly established Network Data Steering Group, which includes not only member state regulators and European Commission experts, but also representatives from patients, healthcare professionals, health technology assessment bodies, payers, and ethicists.

A key achievement is the Data Analysis and Real World Interrogation Network (DARWIN EU), which now provides access to data from 40 different data partners representing more than 290 million patients from 18 EU member states. To date, 79 studies have been completed through DARWIN EU, many of which have increased understanding of clinical context and supported regulatory decision-making. For example, a study on the risk of suicidality following exposure to doxycycline, a treatment for acne, found no increased risk and enabled the European Medicines Agency (EMA) to reassure patients.

The EMRN has also launched the Catalogues of real-world data sources and studies, which promote data and study transparency. By the end of 2025, these catalogues contained almost 300 data sources and over 3,000 real-world data studies, allowing medicines regulators, researchers, and pharmaceutical companies to easily identify and use the most suitable data sources when investigating the use, safety, and effectiveness of medicines.

The transition to data-driven medicines regulation demonstrates how AI governance can work in practice when paired with strong data governance frameworks, quality standards, and multi-stakeholder oversight. As the UN Global Dialogue on AI Governance moves forward, these sectoral examples may provide templates for how international AI governance can function across different domains and jurisdictions.

" }