António Guterres called for mandatory safety testing on AI systems, proposed a child protection pact, demanded transparency on the environmental footprint of data centers, and warned about autonomous weapons and deepfakes. Nothing was signed. Nothing turned out binding. The Global Dialogue on AI Governance convened in Geneva under the United Nations ended exactly as it began: with moral appeals floating without any circuit connecting warning to consequence.
AI governance without an enforcement mechanism is noise dressed up as signal, because it emits a message that is never measured, never corrected, and never generates sanction when ignored. In systems theory this is elementary: real control needs a sensor to detect deviation and an actuator to correct it. Geneva had a microphone but lacked the other two components. Much was said. Nothing was verified.
This tendency didn't start with AI. It appeared in 1944, when forty-four countries met in New Hampshire to decide how the world would organize money after the war.
Bretton Woods achieved something concrete at first: fixed exchange rates, the International Monetary Fund, the World Bank. That structure coordinated expectations for decades. Over time the dominant powers discovered they could attend the forum without ceding effective control. The United States unilaterally abandoned the gold-dollar standard in 1971, and the rest of the system adjusted to that decision. The forum coordinated rhetoric. Real power stayed exactly where it had always been.
The same regularity repeats across other postwar spaces. The Intergovernmental Panel on Climate Change publishes reports of remarkable scientific precision and lacks any tool to sanction a country that fails to meet its targets. The World Trade Organization arbitrates trade disputes, but when a major power ignores a ruling, the entire mechanism seizes up. In each case the moral summons is strong and the enforcement capacity is nil or nearly so.
Why do governments and companies keep attending these forums if they have no intention of ceding anything real?
The answer is that participation itself generates gain. The leading labs and the G7 and OECD governments that set the safety agenda come away with legitimacy. They can claim they supported transparency and child protection. That statement costs zero in terms of concrete change and is worth a great deal in public perception. It serves as a shield against tougher external regulations.
What actually defines what counts as risk in these dialogues? The countries that already dominate the technical frontier. While the United States and China converge on state control of their most advanced models, everyone else debates general principles. Frontier decisions happen in bilateral conversations between labs and defense ministries. The safety agenda tends to focus on distant scenarios and sidesteps the more immediate question: what happens to the workforce in economies without equivalent infrastructure or bargaining power.
This gap stops being symbolic and becomes material. The International Monetary Fund and the International Labour Organization agree that AI exposure is greater in advanced economies. The finding seems reassuring until examined closely: the portion of emerging economies that is exposed corresponds to offshored labor, the most vulnerable and least protected kind. The World Bank estimates that in Latin America seventeen million people will be unable to capture AI benefits due to lack of basic infrastructure. While the communiqués talk about environmental transparency, that divide widens every day without any process recording or correcting it.
The narrative of inclusive governance has genuine value as a starting point. It can function at the same time as cover for inaction. The two things aren't mutually exclusive. It doesn't take bad faith for the result to be institutional theater: it's enough for the design to lack feedback loops. Intent doesn't alter the physics of the system.
I recognize the limits of this critique. Coordinating expectations has real value, and Bretton Woods, for all its subsequent erosion, avoided immediate monetary chaos. The question isn't whether these forums are useless, but whether they're built to correct themselves when they fail. So far the answer remains no.
The 2027 session is the window that deserves close attention. If emerging economies arrive with their own agenda centered on technology transfer, distributed compute capacity, and verifiable compliance processes, the forum could stop being pure ritual. Coordination among the expanded BRICS bloc, the African Union, and middle-income economies in Latin America could generate enough weight to demand something more than principles. I keep exploring these dynamics at Las Piedras No Mienten, and I still don't know whether that moment will arrive or whether the pattern will simply repeat itself.
Göbekli Tepe shows that communities sustained enormous cooperative projects across generations without any central state imposing anything. At some point they decided to deliberately bury part of what they'd built. Coordination without coercion is possible, but it demands a design different from the one Geneva produces today: one where prestige and reciprocity generate real obligation and not just a press release.
What structure would allow global AI governance to incorporate genuine sensors and actuators instead of just well-intentioned declarations?
