The US government has ordered Anthropic to block access to its most powerful AI models. The trigger was apparently a tip from Amazon boss Andy Jassy about a security gap. There are now also indications of Chinese access to one of the models. However, Anthropic criticizes the US government’s actions. For Europe, the incident reveals an uncomfortable truth: Anyone who is dependent on US providers for AI can lose access overnight. A commentary analysis.
Why the US government made Anthropic shut down its AI models
- Following an order from the US government, Anthropic has cut off access to its new top AI models. The company said in a statement that authorities had ordered that access to the Fable 5 and Mythos 5 models be prevented for all non-Americans. The reason was unspecified Threat to national security referred. Since it was not technically possible to block access to individual nations, the company pulled the plug worldwide. According to Anthropic, the government has become aware of a method that can be used to circumvent restrictions in AI software.
- As the Wall Street Journal reports, a phone call is said to have triggered the ban. Accordingly, Amazon boss Andy Jassy called US Treasury Secretary Scott Bessent and other government representatives and pointed out a technical problem. Researchers at the company are said to have discovered that a certain sequence of prompts could cause Anthropic’s AI models to reveal security-relevant information. What was apparently initially intended as a warning triggered a chain reaction. US President Donald Trump apparently even personally approved the measure – despite internal concerns about economic damage.
- Just a few days after the lockdown was ordered, further details came to light. As the US magazine Semafor reports, citing an insider, it is not just the alleged jailbreak that is behind the US government’s order. Rather, there are indications that one Group with ties to the Chinese government had access to Claude Mythos. An official confirmation is pending. However, that would explain why two AI models were banned. The previously confirmed security hole is said to only affect Fable 5.
What the Anthropic lockdown means for Europe
The US ban on Anthropic acts like a geopolitical stress test for Europe’s digital policy. Unfortunately he is no exercise. When presenting its Mythos model, Anthropic itself pointed out that its capabilities could be misused, for example to detect security gaps in software. But what began as a well-intentioned warning about real risks has become an argument for a global access ban.
The question is not necessarily how dangerous the technology is, but rather who gets to decide who can use it. For Europe the process sets an unpleasant dependency open. Because: For the most powerful AI models, the crucial infrastructure predominantly comes from the USA. But: If a political decision in Washington is enough to deprive researchers, companies or authorities in Europe of access to important tools, a technology issue suddenly becomes a question of sovereignty.
The longing for our own top-of-the-line European models is therefore not just an industrial policy vision, but also the desire sober damage limitation. There is therefore growing concern in Brussels that not only AI systems, but also cloud services, chips and digital infrastructures could be used as geopolitical leverage in the future.
The planned European sovereignty package is therefore more than just another strategy paper. The actual test will be whether Europe acts faster this time when analyzing new dependencies. Because digital independence cannot be downloaded. Someone has to develop them.
Voices
- After Anthropic complained about the US government’s actions due to a lack of transparency and too short a deadline for action, Anthropic responded Pentagon spokeswoman Kirsten Davies in a post on
- Ralf Wintergerst, President of the digital association Bitkomin a statement: “Germany and Europe are dependent on the goodwill of the US government for access to the strongest AI models, as the surprising order at the weekend made more than clear. This has a direct impact on the performance of our traditional industries and our administrations and also affects our security and, for example, the excellence of our science. Now more than ever, it must be about making Germany and Europe digitally sovereign. Digital sovereignty and our own AI competencies belong at the top of the political priorities.”
- Also Rasmus Rothe, Chairman of the Federal AI Associationtakes a similar note in a LinkedIn statement: “What we experienced yesterday is no surprise, it is the receipt for years of hesitation. The US government has proven what we have been saying since 2021: AI is strategic infrastructure. Those who do not control it themselves will be controlled by others. We are not demanding another strategy process. We are demanding a summit. Now.”
IPOs, Geopolitics and Sovereignty: What’s at Stake
For Anthropic, the US government’s order comes as one extremely bad timing. The company filed confidential documents for an IPO just a few weeks ago and now has to justify itself to its investors and explain why two of its most important AI models disappeared from the market overnight.
Anthropic itself is currently assessing the alleged trigger for the ban significantly less dramatic than the US authorities. It was only a limited possibility of using AI to analyze and correct certain program codes. It is questionable whether this actually created or could have created a national security threat.
At the same time, Trump’s order is part of a longer history of conflict. Because Anthropic had repeatedly objected to the use of its systems autonomous weapon systems and for mass surveillance and thereby created powerful opponents in Washington. There is currently no reliable evidence that the current ban is politically motivated. But the case shows how closely economic interests, security issues and government intervention are intertwined in the AI age.
In other words: whoever controls the models increasingly also controls the rules of the game. The Consequences also reach far beyond Anthropic. Companies, research institutions and administrations are increasingly building their processes on powerful AI systems. If access is suddenly lost, not only technical but also economic damage will occur – for both domestic and foreign users.
That’s exactly why The EU Commission is monitoring developments closely and reiterates the need for technological sovereignty. The lesson could be that AI is no longer just ordinary software. It is developing into a strategic infrastructure – and thus into a resource that can be just as hotly contested in times of crisis as energy, raw materials or trade routes.
Also interesting: