In mid-June, the U.S. government ordered the blocking of access to Anthropic’s latest frontier models – Claude Fable 5 and Mythos 5 – outside American borders, citing national security risks. The decision signals that the most advanced AI systems are now treated as strategic assets, comparable to cutting-edge semiconductors. For Europe and Poland, this move reveals a dangerous dependency on foreign technology and raises urgent questions about digital sovereignty.
The ban covers foreign governments, companies, and individuals. It is not an isolated incident: the Trump administration has been in conflict with Anthropic since 2025, after CEO Dario Amodei criticized a flagship AI project and opposed the repeal of Joe Biden’s executive order on AI. Although Anthropic signed a Pentagon contract in July 2025 to build prototype systems for national security, the deal was terminated in February 2026 when the Department of Defense demanded renegotiation to remove the company’s restrictions. That dispute ended with sanctions barring Pentagon contractors from doing business with Anthropic.
Behind the official narrative of national security lies a broader geopolitical rivalry, especially with China. As one expert notes, the U.S. is not only worried about the 'offensive’ capabilities of these models but also about protecting its competitive edge and intellectual property. 'Initially, the restrictions targeted chips and infrastructure. When China and others found workarounds – optimizing model efficiency – the U.S. expanded controls to software,’ says Artur Józefiak, vice president and head of the Cybersecurity Services Center for Europe at Accenture.
A strategic shift: AI as a dual-use asset
Karol Paciorek, head of CSIRT KNF, highlights a critical lesson for the financial sector and beyond. 'This decision shows that advanced AI models are beginning to be treated as technologies of strategic importance. The discussion is no longer just about their capabilities, but also about their potential impact on national security, economies, and cyberspace.’ He warns that dependence on a single provider or jurisdiction introduces serious operational risk. 'Entities using such solutions should account for the scenario where access is cut off in their risk analysis,’ he adds.
Józefiak reinforces this: 'We should map our dependencies – what are the digital resources our companies and institutions rely on? Then we need to diversify them, geographically and technologically, where possible. It’s not about abandoning American cloud services overnight, but about building minimal critical capacity so that organizations can function even if part of the supply is severed.’ He suggests that funds from programs like SAFE could be allocated to this dimension of 'digital resilience.’
What this means for Poland and Europe
Europe, often seen as lagging behind due to heavy regulation, faces a stark choice. Hans de Vries, director of cybersecurity and operations at ENISA, points out that AI is already reshaping the threat landscape. 'Large language models are used to enhance phishing and automate social engineering. Frontier AI models compress the vulnerability management lifecycle, finding zero-day exploits faster than ever.’ He stresses that 'security on paper’ no longer suffices – organizations must move from strategy to execution, adopting zero-trust principles and defenses that operate at machine speed.
The situation is evolving rapidly. On the same week, CNN reported that the White House asked OpenAI to restrict access to its upcoming GPT 5.6 model, limiting it to government-approved partners. This pattern suggests that the U.S. intends to treat any software deemed strategic as a tool for diplomatic and military leverage.
For Poland, the risks are concrete. Polish financial institutions, government agencies, and companies using American AI models may suddenly find themselves cut off or forced to comply with U.S. policy conditions. The solution, as Józefiak concludes, is not to create national alternatives out of thin air, but to understand where the dependencies lie, spread them out, and build a bare minimum of own capabilities in critical areas. 'Digital resilience is not just about AI’s offensive capabilities. It’s about the entire layer of digital dependencies that, for geopolitical reasons or because of disruptive technologies, could paralyze a country’s functions.’
Źródło: WNP.PL, Fot. Shutterstock
