A US decision to cut Europe off from Anthropic’s most powerful AI models has sparked a backlash in Brussels, with the Commission warning against discriminatory restrictions and lawmakers arguing the episode proves Washington holds a technological “kill-switch” over Europe.
The restrictions have reignited concerns about Europe’s dependence on American technology at a time when artificial intelligence is increasingly viewed as a strategic asset for cybersecurity, economic competitiveness and national security. Brussels insists any company operating in Europe must comply with EU rules. Members of the European Parliament say the bloc can no longer rely on access that a foreign government can withdraw overnight.
“We’re ready to engage and tackle these security risks together with like-minded partners,” European Commission spokesperson Thomas Regnier said, pointing to the AI Act and the EU’s cyber legislation as existing safeguards.
“But any company, no matter what, has to comply with our legislation if at some point their service or model is offered in Europe — be it the AI Act or our cyber legislation,” he added. Regnier added that discussions with tech CEOs would take place on the sidelines of the G7 summit. However, he did not confirm whether Anthropic’s restrictions would be raised.
You might be interested
Washington’s decision came after the Trump administration ordered Anthropic to cut off its two most powerful AI models from all non-American users, citing concerns that the technology could be exploited by adversaries such as China and Russia.
Fable 5, the company’s latest commercial model, was blocked internationally within days of its launch. The restrictions also affect Mythos 5, an even more capable underlying model available only through Anthropic’s cybersecurity research programme, Project Glasswing.
The models are designed to autonomously identify software vulnerabilities at a scale and speed beyond human researchers, making them valuable tools for cyber defence but also potentially dangerous if misused. Anthropic says Mythos 5 has discovered more than 10,000 high- and critical-severity software vulnerabilities, including zero-days that had survived decades of human review.
Project Glasswing opens — then closes — Europe’s door
The restrictions are particularly striking because they come only weeks after the EU’s cybersecurity agency, ENISA, became the first European organisation admitted to Project Glasswing, with NATO added in the same expansion.
Launched in April, Project Glasswing is Anthropic’s cybersecurity coalition for trusted security teams. It gives them access to the latest AI models before public release. Initial partners included Amazon Web Services, Apple, Google, Microsoft, Cisco, NVIDIA and the Linux Foundation
Not everyone, however, is impressed by the hype. Bruce Schneier, cryptographer and lecturer at Harvard University, said that “Fable is just another incremental improvement in the years-long climb of AI capabilities.”
“Those with access used Mythos to find and patch many vulnerabilities in their own software. But one UK group found the latest, already public, OpenAI model to be just as powerful. With Mythos demonstrating that it was possible, the open-source community scrambled to build harnesses that could steer other AI models towards similar capabilities. They largely succeeded. For example, a Prague company was able to replicate Anthropic’s few verifiable cybersecurity capabilities with a much smaller and cheaper model — and a more sophisticated harness,” he wrote in an op-ed.
MEPs warn of digital dependence
Most of the response from MEPs from across the EU focused on the need for “tech sovereignty” and “home-grown solutions”, framing the move as a decision by President Trump to deliberately restrict European access to Fable 5 and Mythos 5.
Bart Groothuis (Renew/NLD) sits on the Committee on Industry, Research and Energy and negotiated the current Chips Act. He said: “These restrictions are a clear example of the current American ‘nobody but us’ mentality. While software is vulnerable worldwide, the mitigation of potential hazard is left to Americans only. It’s irresponsible. With the ascent of frontier AI models, we have a huge task ahead of us to avoid a cyber bloodbath. Once again: this shows that Europe needs its own LLMs and open weight models or face digital colonisation.”
The US holds a real ‘kill-switch’ over essential technologies and they are more than willing to use it.
— Christophe Grudler, MEP, Renew Europe (FRA)
Christophe Grudler (Renew/FRA) coordinates the Committee on Industry, Research and Energy. He added: “The United States is once again demonstrating what we Liberals and Democrats have warned about so many times since Trump entered into office; that the US holds a real ‘kill-switch’ over essential technologies and that they are more than willing to use it. Europe needs to put on the turbo in its strive to reach digital independence. The only way to reach it is by mobilising billions of euros in our own European AI ecosystem.”
Aura Salla (EPP/FIN) said: “Europe cannot keep building its tech stack on access that can be switched off overnight by a foreign government. We must take action to reserve our data and our market primarily for European tech to scale it and build our own frontier AI.”
