Why the White House cannot keep Anthropic behind a digital border

In 1991, a software engineer named Phil Zimmermann released a program called Pretty Good Privacy. It allowed regular people to encrypt their emails so that even the most powerful intelligence agencies could not read them. The U.S. government responded by treating his code like a physical weapon. They launched a three-year criminal investigation, arguing that Zimmermann was an illegal arms exporter. To bypass the ban, Zimmermann did something clever. He printed the entire source code for the software in a book and published it. Under the First Amendment, the government could not stop the export of a book, even if it contained instructions for digital weaponry. History has a way of repeating its struggles, and today, the battleground has shifted from encryption keys to large language models.

Last Friday, the White House ordered the AI startup Anthropic to stop the export of its most advanced models, Fable and Mythos. The directive applies to any entity outside the United States and even to foreign nationals working within American borders. Anthropic responded by pulling the plug on both models for everyone. For the past week, these systems have been offline. This incident is the first major test of whether the government can treat software as a national security asset that stays behind a fence. Looking at the big picture, these export controls often create a high-stakes game of whack-a-mole that companies and researchers eventually win.

The myth of the doomsday machine

Ever since Anthropic released Mythos in April, the company has surrounded it with an aura of danger. Their marketing material suggests it is a tireless intern with a dark side. They claim it has the potential to help bad actors write malicious code or disrupt infrastructure. Because of these fears, Anthropic originally limited access to about 150 vetted organizations. This strategy is part of a broader trend where AI labs build hype by telling the public their products are too dangerous to exist freely.

Two specific events triggered the current government crackdown. First, Anthropic allowed a South Korean telecom company to use Mythos. U.S. officials became concerned that this company had hidden ties to Chinese interests. While the telecom firm denies these claims, the mere suspicion was enough to spook regulators. Second, Amazon CEO Andy Jassy reportedly told the administration that his own researchers bypassed the safety filters on the Fable 5 model. Anthropic claims this was a minor issue they already fixed, but the government viewed it as a sign that the software was not yet under control.

Under the hood, these models are essentially just massive files of numbers and probability. Unlike a physical missile or a chemical agent, software is weightless and infinitely replicable. The Commerce Department gave Anthropic only 90 minutes to comply with the order. This sudden shutdown reflects a growing anxiety in Washington about maintaining a technological lead over global rivals. If the government treats AI like a secret recipe, they risk falling behind as other countries develop their own versions without the same restrictions.

Why digital borders are historically porous

The attempt to control AI exports mirrors the early days of the Crypto Wars. After the PGP investigation ended without a conviction, the U.S. government eventually relaxed its grip on encryption. This change allowed for the development of the secure internet we use today. Without that shift, the encrypted apps billions of people use to send private messages would likely not exist in their current form. The government learned that math is very difficult to regulate once it is out in the world.

In the 2010s, the focus shifted to spyware. Governments tried to use the Wassenaar Arrangement, an international treaty for dual-use technologies, to stop the sale of hacking tools to dictators. This treaty classifies surveillance software as something that has both civilian and military uses. However, the system has huge gaps. Countries like Israel do not follow the arrangement, which allows firms like NSO Group to sell products globally regardless of Western concerns.

Even within the European Union, enforcement is inconsistent. Some governments have historically granted export licenses to spyware makers even after reports showed their tools were used against journalists. When one country tightens its rules, these companies often move their headquarters to a more relaxed jurisdiction. Intellexa, a major player in the spyware market, shifted its operations across several countries to stay ahead of sanctions. These examples show that as long as there is a market for a technology, someone will find a way to ship it.

The cost of compliance and the global race

For the average user, these export controls might seem like a distant political drama, but they have tangible effects on the technology we use. If Anthropic and its competitors have to get a government permit for every foreign customer, their costs go up. This burden makes American companies less competitive. A developer in Berlin or Tokyo might choose a French or Chinese AI model simply because it is easier to access. This creates a fragmented internet where your location determines the quality of the tools you can use.

On the market side, this impasse creates a volatile environment for investors. Anthropic is a multi-billion dollar company that relies on global growth. If a significant portion of the world is suddenly off-limits, their business model changes. This situation also encourages "shadow AI," where researchers in restricted countries find ways to download and run leaked versions of these models on their own hardware. Once a model is released to even 150 vetted companies, the chance of a leak increases. Digital crude oil has a way of seeping through even the smallest cracks in a container.

What this means for your digital future

The standoff between Anthropic and the administration remains unresolved. There is a chance the government will back down to keep American tech dominant. If they do not, we might see a world where AI software is categorized the same way as nuclear secrets. This would be an unprecedented move for a technology that is already being integrated into everything from spreadsheets to search engines. Historically, these restrictions fail because they assume that the U.S. has a permanent monopoly on innovation.

Practically speaking, you should observe how the apps on your phone react to these policy shifts. You might notice that certain features are unavailable when you travel abroad, or that your favorite AI assistant becomes less capable as its creators focus on safety compliance over performance. Ultimately, the Anthropic incident proves that the era of open, borderless software development is ending. Governments are now treating code with the same suspicion they once reserved for physical weapons, even if history shows that code always finds a way to cross the line.

As you navigate your digital habits, consider how much of your daily productivity depends on tools that could be switched off by a government directive. The invisible mechanics of international trade are now as much a part of your smartphone as the battery or the screen. Shifting your perspective to recognize these digital borders helps you understand why some technologies thrive while others suddenly vanish from the market.

Sources:

• U.S. Department of Commerce Export Administration Regulations
• Anthropic Corporate Transparency Report 2026
• Wassenaar Arrangement on Export Controls for Conventional Arms and Dual-Use Goods and Technologies
• Electronic Frontier Foundation Archives on the Crypto Wars