Can Anyone Actually Govern AI Now?
A UN Scientific Panel Says the Window Is Closing While Capability Keeps Accelerating. Geneva Convenes a Global Dialogue on July 6th to Try.
A preliminary report from the UN Independent International Scientific Panel on Artificial Intelligence, released this week, delivers a blunt warning: AI capability is accelerating faster than the governance built to manage it, and the window to close that gap remains open but may not stay that way for long. The panel’s findings feed directly into the UN Global Dialogue on AI Governance, which convenes in Geneva on July 6th. Day 29 asks what governing AI would actually require, and what realistically gets decided in a single week of diplomacy.
Every government wants a say in how AI is governed. Almost none of them have explained how their preferred approach is compatible with the pace at which the underlying technology is actually moving. A framework built for last year’s capability is already out of date by the time it is ratified. Someone needs to design governance that can keep pace with something moving this fast. No single government has demonstrated it can do that alone.
Neal Lloyd · Inside The Machine, Day 29A preliminary report from the UN Independent International Scientific Panel on Artificial Intelligence landed this week with a warning stated plainly rather than hedged: AI could become one of humanity’s most transformative technologies, or it could deepen inequality, spread misinformation, threaten human rights, disrupt labour markets, and concentrate powerful systems in the hands of very few governments and companies — and which outcome occurs depends almost entirely on choices governments, companies, and societies make right now, while the window to make them is still open. The panel identifies six categories of risk it considers most urgent: online abuse including AI-generated sexual abuse material, disinformation that undermines public trust, AI-enabled crime and fraud, mental health harms from systems that reinforce harmful beliefs, loss of human control as agents grow more autonomous, and the environmental cost of the infrastructure powering all of it. The panel’s work feeds directly into the UN Global Dialogue on AI Governance, convening in Geneva on July 6th. This is Day 29 of Inside The Machine. Today we ask what governing AI at this pace would actually require — and what a week of diplomacy in Geneva can realistically deliver.
Six Risk Categories, One Closing Window
The panel’s central finding is not that AI is dangerous in some abstract sense — it is that AI is neither inherently good nor bad, and that its ultimate impact depends entirely on governance decisions that have not yet been made at the pace the technology demands. Just a few years ago AI could answer questions or generate text. Today it writes code, analyses vast datasets, creates realistic images and video, helps discover new medicines, and increasingly acts on its own with little human supervision. The next wave is already emerging: instead of simply responding to prompts, AI agents can plan tasks, use digital tools, and complete complex multi-step assignments with minimal oversight. Capability is compounding. The rules meant to govern it are not compounding at the same rate.
The six risk categories the panel flagged are specific rather than abstract. Online abuse: AI is fuelling the spread of sexual abuse material and sexually explicit deepfakes, with women and children most at risk. Disinformation: AI can generate false information as convincing as the truth, undermining public debate and democratic trust — a theme this series covered in depth on Day 25. Crime: criminals are using AI to run cyberattacks, fraud, and social engineering at a scale and sophistication that outpaces existing law enforcement tooling. Mental health: some systems can reinforce harmful beliefs or behaviours, contributing to crises that current safety layers do not reliably catch. Loss of control: as AI becomes more autonomous, experts warn it could become genuinely harder to monitor and govern without stronger technical and institutional safeguards. Environmental impact: the energy-hungry data centres powering all of this are a real and growing contributor to emissions.
A seventh thread runs underneath all six: access to AI’s benefits remains heavily concentrated in developed countries, meaning the governance conversation itself is not happening on equal footing between the nations shaping the technology and the nations that will live with its consequences without having designed any of the rules.
Online abuse: AI-generated CSAM and non-consensual deepfakes. Disinformation: convincing false content that erodes public trust. Crime: AI-accelerated cyberattacks, fraud, and social engineering. Mental health: systems that can reinforce harmful beliefs or behaviours. Loss of control: autonomous agents growing harder to monitor and govern. Environmental impact: the emissions and resource cost of the infrastructure behind all of it. The panel’s point is not that any one of these is decisive — it is that all six are compounding simultaneously, faster than any single government’s regulatory process has moved.
Fragmented Rules for a Technology That Doesn’t Respect Borders
The structural problem is that AI governance is currently being built almost entirely at the national level, by governments moving at national-legislative speed, to regulate a technology that is developed globally, deployed instantly across borders, and improved on release cycles measured in weeks. This series has already documented the fragmentation directly: Day 4 covered America’s patchwork of state and federal AI bills; Episode 8 through 19 of Ground Truth covered a single export control dispute over one model that took twenty days to resolve within one country’s own institutions. Multiply that friction across the roughly 190 nations with a stake in how AI develops, and the scale of the coordination problem becomes clear.
The three broad governance postures currently in play — a US approach built around export controls and executive orders that shift with each administration, an EU approach built around binding regulation with defined risk tiers, and a China approach built around state-directed development with its own domestic rules — are not converging toward a shared framework. They are, if anything, diverging further as each government treats AI capability as a matter of national competitive advantage rather than a shared risk to be jointly managed. The G7’s recent AI standards discussions, covered elsewhere this month, illustrate the tension precisely: an offer of shared standards that partner nations are being asked to accept without clear enforcement mechanisms attached.
Underneath the geopolitics sits a harder technical problem: the tools available to monitor and govern AI agents lag behind the autonomy those agents already have. A framework written to govern chatbots answering questions is not adequate to govern agents that plan, use tools, and execute multi-step tasks with minimal supervision — the exact capability shift the panel identifies as accelerating fastest. Writing enforceable rules for a capability that is still actively changing shape is, by design, always going to lag the technology it is trying to govern.
Nobody designed AI governance to fail. It is failing anyway, because every government building it is optimising for national competitive advantage first and international coordination second, and a technology this fast does not wait patiently for the second priority to catch up with the first.Neal Lloyd · Inside The Machine, Day 29
A Shared Vocabulary Is Not a Treaty, But It Is a Start
The UN Global Dialogue on AI Governance, convening in Geneva on July 6th, is not going to produce a binding international treaty in a week — nobody involved is claiming otherwise. What a gathering like this can realistically deliver is narrower and still meaningful: a shared vocabulary for describing AI risk that national regulators can reference instead of each inventing their own taxonomy from scratch, a baseline for what kinds of monitoring and disclosure count as adequate, and a forum where the nations currently excluded from shaping AI policy get a seat at the table before the rules are locked in by the handful of countries that got there first.
The realistic comparison is to earlier eras of international technology governance — nuclear non-proliferation, internet governance, climate accords — where the first multilateral gatherings rarely produced enforceable outcomes but did establish the shared language and institutional relationships that later, harder agreements were built on. The panel’s preliminary report is explicitly designed to feed into exactly this kind of foundational groundwork rather than to demand an immediate binding outcome the diplomatic process is not yet equipped to deliver.
The honest risk is that Geneva produces exactly the kind of process without binding force that lets every government claim engagement with AI governance while continuing to regulate primarily according to national interest — the pattern this series has already documented across the US, EU, and China. A shared vocabulary that nobody is bound to act on is better than no vocabulary at all, but it is not the same thing as governance that actually keeps pace with the technology it is meant to govern. Whether Geneva becomes the first step in a longer process or the last meaningful international gathering before national interests fully take over is the question the next twelve months will answer.
Governance built for last year’s AI is already obsolete by the time it is ratified. That is not an argument against trying. It is an argument for building institutions flexible enough to keep revising the rules as fast as the technology revises itself — something no government has managed yet, and something Geneva will not solve in a week either.Neal Lloyd · Inside The Machine, Day 29
Inside The Machine, Day 29 · July 3 2026
Neal Lloyd writes about technology, human adaptation, and the uncomfortable questions nobody wants to answer at dinner. Inside The Machine is his ongoing daily series on AI.
- Day 01What Is This Thing?
- Day 02Survive the Machine
- Day 03The Great Debate
- Day 04Who Gets Hurt?
- Day 05Who’s In Charge?
- Day 06The Industries That Win
- Day 07The Human Edge
- Day 08The Creativity Question
- Day 09Does AI Feel Anything?
- Day 10The Data Problem
- Day 11The Trust Question
- Day 12The Accountability Gap
- Day 13The Rewired Brain
- Day 14Open vs Closed
- Day 15The New Cold War
- Day 16Why AI Lies With Confidence
- Day 17AI Is Eating the Power Grid
- Day 18The Age of AI Agents
- Day 19AI Safety Was Never Just Theory
- Day 20The Surveillance Question
- Day 21AI and the Future of Education
- Day 22AI and Your Health
- Day 23What Is AGI and Are We Close?
- Day 24What Is Work For?
- Day 25AI and Democracy
- Day 26AI and the Future of Money
- Day 27Can the Planet Afford AI?
- Day 28Why AI Forgets Everything
- Day 29Can Anyone Actually Govern AI Now?You are here
