Anthropic says the U.S. government ordered it to suspend access to Fable 5 and Mythos 5 for foreign nationals. The company says the order is broad enough that it had to turn the models off for everyone.
Maybe there is a real security reason behind it. Maybe there is a technical risk Anthropic cannot talk about. Maybe this is industrial policy wearing a security badge. From the outside, we don't know.
But the result is easy to understand: someone in Washington moves a lever, and people outside the United States lose access to one of the most advanced AI systems available.
The dependency is personal
I live on this digital highway. I work with these tools. I test them, use them, connect them to business systems, and build workflows around them. For me they are not toys or a headline on a tech site. They are part of the infrastructure of modern software work.
So when the United States closes a door, I feel it. When China controls a large part of the hardware supply chain, I feel it. And when Europe enters the AI race with a brake pad jammed into the wheel, I feel that too.
That last part is the one that annoys me most.
Regulation is not construction
Europe has spent years talking about digital sovereignty. Too often, it treats sovereignty as a synonym for regulation. Regulation is not construction. Regulation does not raise capital. Regulation does not buy GPUs. Regulation does not ship models. Regulation does not keep a startup alive while it burns cash trying to compete with an American lab backed by billions.
Rules matter. I don't want an AI race with no responsibility and no limits. But there is a difference between guardrails and a wall in front of the car.
The United States has OpenAI, Anthropic, Google DeepMind, Meta, NVIDIA, Microsoft, Amazon, and a capital market that can fund absurd levels of risk. China has its own models, its own hardware strategy, its own protected market, and a state that knows this is a strategic race.
One good card is not a strategy
Europe has Mistral.
Thank God for Mistral. I mean that seriously. It is one of the few European AI companies that can sit at this table without sounding like a policy paper pretending to be a product. But if our answer to an Anthropic shutdown is “at least we have Mistral,” that tells us how thin our position is.
The United States has a deck of cards. China has a deck of cards. Europe has one good card and a stack of compliance documents.
The talent is here. The system is heavy.
This is not a talent problem. Europe has talent. The problem is that we ask that talent to run uphill with stones in its backpack.
Raising money is harder. Scaling is harder. Selling to public institutions is slower. Hiring across borders is more bureaucratic. Compute is expensive. Legal risk arrives early. Reputational fear arrives even earlier. Before a company can prove it can compete, it is already being asked to survive a maze built by people who have never shipped a product under pressure.
That is not how you win a technology race.
Then we act surprised when founders leave. We act surprised when European products are born with one eye on the U.S. market. We act surprised when our companies depend on American APIs, Asian chips, American cloud platforms, and foreign capital.
Digital independence is not declared. It is paid for. It is built. It is bought with risk.
The warning is not Fable 5
The Anthropic case should sting because it shows the dependency clearly. Fable 5 itself is not the whole story. Tomorrow we will use another model. Work will continue.
The problem is that access to a technology that will shape productivity, defense, education, science, software, and business can be changed by a political decision made somewhere else.
If that does not bother Europe, then Europe has not understood the game.
What Europe should do instead
The answer cannot be another speech about safer innovation. It has to be money, procurement, infrastructure, serious support for founders, and a tolerance for failure that Europe still does not have.
Stop treating every new capability as a crime scene waiting for an investigator.
Right now the balance is ugly: the United States can switch off models, China can restrict hardware, and Europe can publish a compliance guide.
That is not sovereignty.
That is dependency with nice typography.