The Best AI in the World Is Useless If You Can't Access It
The US has already restricted access to frontier models. Now China is considering controls of its own.
Published:
We spend an enormous amount of time arguing about which AI model is best.
Is OpenAI ahead of Anthropic?
How close are Chinese models to the American frontier?
Will closed models win, or will open-weight models make the whole race irrelevant?
Every new benchmark seems to produce a new winner.
But the events of the last few weeks suggest that there is another question we may need to ask first:
Will you still be allowed to use that model tomorrow?
A model can be extraordinarily capable and still be useless to you if access from your country is blocked, if a government delays its release, if your nationality makes you ineligible, or if the provider is required to change who can use it.
The model can continue to exist. The company can remain healthy. The servers can keep running. Access can still disappear.
Frontier AI access is beginning to look less like an ordinary software market and more like a matter of national policy.
The First Warning Came from Anthropic
On June 12, the US government applied export controls to Anthropic's Claude Fable 5 and Claude Mythos 5. According to Anthropic, the directive required the company to prevent foreign nationals from accessing the models, whether those users were inside or outside the United States. Anthropic said it had no reliable way to verify nationality in real time, so it suspended access to both models for everyone. [1]
The model did not disappear.
The company did not shut down.
The infrastructure did not fail.
Policy changed, and access disappeared.
The controls were later lifted. Anthropic announced on June 30 that Fable 5 would return globally on July 1, while Mythos 5 access was restored to a set of US organizations and remained subject to further coordination. [2]
It is easy to look at the final outcome and say that the interruption was temporary. That misses the more important point.
We have now seen a frontier model remain technically available while policy decisions determined whether users could reach it. That is no longer a hypothetical risk.
OpenAI Showed Another Version of the Same Problem
OpenAI's GPT-5.6 rollout followed a different path, but the underlying shift was similar.
When OpenAI announced the GPT-5.6 family on June 26, it did not begin with broad public availability. OpenAI said it had previewed the models and its release plans to the US government and, at the government's request, started with a limited preview for a small group of trusted partners. The company also said that the participating organizations had been shared with the government. [3]
OpenAI explicitly said that it did not want this kind of government-access process to become the long-term default. It still expected broader availability and continued testing during the preview period. [3]
On July 8, Reuters reported that the broader GPT-5.6 rollout was set to proceed after additional testing and discussions between OpenAI and US officials. [4]
The important point is not merely that GPT-5.6 is moving toward wider release.
The important point is that the timing and scope of a frontier-model release became something discussed with government before the public received broad access.
That does not mean every AI release is now directly controlled by the state. It does mean that national-security policy has moved much closer to the moment when advanced models become available to ordinary users and developers.
Now China Is Considering Controls of Its Own
This is no longer only an American story.
On July 7, Reuters reported that Chinese authorities had held meetings with major technology companies about potentially restricting overseas access to China's most advanced AI models, including models that have not yet been released. Reuters reported that Alibaba, ByteDance, and Z.ai were among the companies that attended the discussions. [5]
According to the report, the talks covered both closed models and more open versions. The possible scope is still being discussed, may apply only to future models, and may never be implemented at all. [5]
China has not closed the door.
But the fact that the door is now being discussed matters.
Until recently, one could still imagine the AI model market moving toward something close to a global software market: American providers competing with Chinese providers, open-weight models crossing borders, and users choosing among them mainly on capability, cost, latency, and product quality.
That assumption is becoming harder to make with confidence.
The United States has already used national-security authorities to restrict access to frontier models. OpenAI has already adjusted a frontier-model rollout after government involvement. China is now reportedly considering whether its own most advanced models should remain freely accessible overseas.
Taken together, these are signs that leading AI systems are increasingly being treated not merely as commercial products, but as strategic national assets.
We May Be Leaving the Era of One Global AI Market
It is too early to say that the AI world will split into national blocs. That outcome is not inevitable, and I hope it does not happen.
But it is no longer difficult to imagine.
One future could include a US-centered model ecosystem, a Chinese model ecosystem, separate European regulatory requirements, and different access rules depending on where a person or organization is located.
In that world, the question "What is the best model?" becomes incomplete.
You also need to ask:
- Can I access it from where I live?
- Can my organization use it?
- Will the API still be available in my region?
- Can I export my data if access changes?
- Can I move my work to another model without rebuilding everything?
- What still works if the provider, policy, or network changes?
Today, choosing one provider because it is cheaper or more capable may be completely rational. The problem is not that users make that choice.
The problem appears when years of memory, work, configuration, and dependence are built around the assumption that the same access conditions will continue indefinitely.
Performance Is Not Enough. AI Resilience Matters Too.
None of this is an argument for abandoning cloud AI.
Frontier cloud models will probably remain more capable than local models for many tasks. They may offer better reasoning, faster coding assistance, stronger multimodal capabilities, larger context windows, and access to infrastructure that most individuals could never run at home.
Use them.
Use the best model available for the work in front of you.
But using a powerful service is different from designing your entire digital life so that nothing survives when access to that service changes.
I think AI resilience may become as important as AI performance.
That means building for the possibility of change:
- models can be replaced;
- providers can be changed;
- data can be exported;
- memory does not exist only inside one company's service;
- a local model can provide a minimum continuity path when cloud access is unavailable;
- the user's accumulated state survives even when the reasoning engine underneath it changes.
A local model does not need to match the best cloud model to be valuable in this design.
Its role can simply be to keep the system alive, preserve access to the user's own state, and provide a path forward until another capability source is available.
Why I Am Building doll
This is one of the reasons I am building doll.
doll is not an attempt to create a foundation model that competes with OpenAI, Anthropic, or Chinese AI labs.
It is a project to keep memory, identity, settings, data, work state, and continuity on the user's side while allowing the underlying AI model to change over time.
The goal is local continuity first, with cloud models added as optional sources of greater capability when they are available and appropriate.
I do not know whether AI will actually divide into national blocs.
I do not know whether future access controls will become broader or more permanent.
I hope they do not.
But after the last few weeks, the possibility is harder to dismiss than it was before.
The most important AI may not be the smartest one.
It may be the one you can still use tomorrow.
Notes and Sources
- Anthropic — Statement on the US government directive to suspend access to Fable 5 and Mythos 5
- Anthropic — Redeploying Claude Fable 5
- OpenAI — Previewing GPT-5.6 Sol: a next-generation model
- Reuters — OpenAI set to launch most capable GPT model after delayed rollout
- Reuters — Beijing is looking at curbing overseas access to China's top AI models, sources say