The AI Standards Race Is Still Open
AI is still young. If open societies throttle their own labs or block global access, they may not slow AI. They may surrender the rulebook.
AI governance is converging on paper while the real standards race is still being fought in code, prices, access and speed.
Global Policy Journal argued this week that AI governance is converging in documents but diverging on the ground. The UN Foundation asked how global AI governance can build a better world. UN News and Global Issues reported that AI systems are still getting women wrong as gender bias persists. The Inter-American Development Bank described AI agents as a new operational layer for digital government.
Those stories point to a hard truth. AI is still young. The world is trying to regulate a technology whose full shape is not yet visible.
That does not mean there should be no rules. It means the first duty of governance is to avoid mistaking fear for strategy.
The danger of premature throttling
AI is powerful, but it is not magic. It is built from imperfect human data, trained inside imperfect institutions and deployed by companies and states with incentives that deserve scrutiny. Bias, error, concentration of power and misuse are real problems.
But the answer cannot be to throttle the technology so heavily that open societies slow their own best builders while others race ahead.
If the United States blocks leading American labs from making frontier models available globally, it may not stop AI diffusion. It may only create a market gap. Developers, governments and companies still need models. If Anthropic, OpenAI, Google or other American firms are restricted from serving much of the world, cheaper and faster competitors will fill the space.
Some will be Chinese. Some will be open-source communities. Some will be firms in countries that care less about democratic oversight than market capture.
The result could be perverse: rules written in Washington that reduce Washington's influence over the very standards the world ends up using.
Whoever ships writes norms
Technology standards are not set only in treaty rooms. They are set by the tools people use.
A model that becomes cheap, reliable and widely available teaches developers what defaults to expect. It shapes APIs, safety practices, evaluation methods, languages supported, moderation assumptions and public-sector procurement. It decides which country's firms train the world's engineers by habit.
If China can ship models that are good enough, cheap enough and available enough, it will not need to win every philosophical argument about AI governance. Adoption will do part of the work. Within a short period, perhaps less than two years if current cost and capability curves hold, Chinese systems could become the default in large parts of the world.
That is a forecast, not a fact. But it is a risk serious enough to change the policy question.
The question is not only, "How do we restrain AI?" It is, "Who writes the operating habits that billions of people inherit?"
Access is a governance tool
Open societies should care about global access not only because companies want revenue. They should care because access is influence.
If democratic firms are present in global markets, they can be pressured by courts, journalists, civil society, shareholders and regulators. They can be required to publish safety practices, submit to audits, respond to bias claims and accept public criticism. Their home governments can be challenged.
Authoritarian systems can also produce useful technology. The issue is not talent. The issue is accountability. If the world's AI layer is built by actors that face weaker public challenge, users may get lower prices and fewer rights at the same time.
That is not a democratic bargain.
Govern without surrendering
A better approach would separate genuine safety rules from self-defeating scarcity.
Governments should require transparency for high-risk systems, redress when automated decisions harm people, testing for bias, clear liability, public-sector procurement standards and strong privacy protection. They should restrict military, surveillance and coercive uses where rights are at stake. They should protect citizens from being governed by black boxes.
But they should be careful about rules that keep democratic models away from the global public while leaving the field open to less accountable rivals.
AI governance should channel the technology toward shared prosperity and democratic consent. It should not become a policy of voluntary retreat.
The standards question
The global AI rulebook will not be written once. It will emerge from law, markets, infrastructure, developer habits and state power. Open societies still have a chance to shape it. They have strong labs, deep capital, universities, civil society, courts and public debate.
But they cannot shape the world from behind a locked gate.
AI is still in its infancy. That is exactly why the next two years matter. The standards formed now may harden before democratic institutions understand what they gave away.
The task is not to worship AI or fear it into paralysis. The task is to build, govern and remain present.
If open societies leave the field, someone else will write the defaults.
The Global Federation treats democratic order as a design problem: rights, duties, institutions and public trust have to be built before crisis hands the future to fear.