AI under lock and key

AI under lock and key

The Anthropic order shows how quickly any system built on a US LLM can be switched off, without notice.

EM
Eduardo Martínez de la Fe
Jun 25, 2026
3 min read

Last week, the US government ordered Anthropic to suspend access to its two most advanced AI models (Claude Fable 5 and Mythos 5) for any foreign national, whether inside or outside the United States. It is the first time export controls have been pushed beyond hardware (chips and semiconductors) and applied directly to an AI software model.

The order arrives in the middle of a deeper confrontation between Anthropic and the Trump administration. Although the official case rests on an algorithmic flaw said to raise national security concerns, Anthropic notes in its own statement that the same vulnerability exists in GPT-5.5 and other rival systems that remain in operation.

Friction with the Pentagon

Earlier this year, Anthropic chief executive Dario Amodei refused to hand over the company’s models to the Department of Defense for use in mass surveillance systems or autonomous weapons without human oversight. Washington responded by classifying the company as a supply-chain risk. A federal judge then blocked that move, describing it as unlawful First Amendment retaliation. The pressure has now returned by another route.

The same method

For months, the Trump administration has used regulatory and commercial power as leverage against those who resist its line. Squeezing Anthropic on the eve of its stock market debut, at a reported valuation of $965 billion, fits a broader pattern of incentives and reprisals that much of the world (including long-standing allies) has already learned to recognise. The tariff threats aimed at European countries that opposed Washington’s plans over Greenland, and the opaque tariff exemptions granted to politically connected companies, belong to the same playbook.

Europe has spent years building its research base, its companies and parts of its public administration on predominantly American AI models. The Anthropic order shows how quickly any system built on a US LLM can be switched off, without notice. It also reads as a warning to the rest of the sector about the price of openly resisting federal demands.

Spain too

This shockwave is also reaching Spain, which has invested €100 million in European digital sovereignty projects. Only weeks ago, Spain’s Ministry for Digital Transformation was welcoming its admission to Project Glasswing, Anthropic’s restricted programme giving selected cybersecurity partners access to Mythos across critical infrastructure in more than fifteen countries. Organisations that had already woven those models into their workflows found themselves without tools within hours, with no warning and no immediate remedy.

What now?

Spain already has nine projects selected for IPCEI-AI, the European programme for sovereign AI development, involving companies such as TelefónicaIndra and Multiverse Computing. At continental level, Mistral remains the only European contender with its own infrastructure, frontier models and data hosted under EU jurisdiction.

The route is there. What is missing is a sense of urgency proportionate to the risk now on display. The European Commission’s recent Tech Sovereignty Package, designed to reduce dependence on foreign cloud, AI and semiconductor providers, cannot remain a statement of intent.

This article is republished from Futuribles. Here's the original article in Spanish and English.

About the Writer

More from Mindplex

Keep reading

Three more ideas worth your time.

Browse Community

Discussion

Join the discussion

Sign in to share a response with the community.

Type @ to mention someone Type / or use + to add a block Highlight text, then choose Link
Loading editor

Comments cannot be edited after posting because they become part of the reputation record. Give yours a quick review first.

Whether this plays out exactly this way or not, it's a good reminder that relying too heavily on a single AI provider can create real strategic risks.

NA

Naod Abebe

2 months ago

This is a massive wake-up call for Europe and the rest of the world.

It is alarming to watch governments attempt to cage AI development using bloated bureaucratic justifications.

Blocking a company's software over a fake "security flaw" while ignoring the exact same issue in other AIs is just a weak excuse to punish Anthropic for refusing to help with mass surveillance. Bullying tech companies like this shows exactly why the rest of the world can't rely on just one country for its AI. If another government can pull the plug on your tools overnight, building your own independent technology isn't just a nice idea anymore, it is an absolute must-have.