Artificial Intelligence & Machine Learning,
Geo-Specific,
Next-Generation Technologies & Secure Development
Decision to Restrict Access Exposes EU Dependency on US Made Models
The decision made by the U.S. government to restrict access for foreign nationals to Anthropic’s cutting-edge artificial intelligence models has ignited a significant debate in Europe about the continent’s reliance on American technological capabilities. The United States’ abrupt implementation of export controls targeting the Mythos and Fable models compelled Anthropic to swiftly discontinue access for a wide range of users, resulting in considerable dissatisfaction across Europe.
The imposed restrictions affect not just foreign nationals operating within the United States, but also those located elsewhere, with even employees of Anthropic impacted. The Mythos model is renowned for its advanced ability to detect and exploit software weaknesses, while the newly introduced Fable was designed with robust cybersecurity measures intended to safeguard its users (see: US Pulls the Plug on Anthropic’s Top AI Models).
Reports indicate that concerns about Fable’s security measures being circumvented were raised with the U.S. government, potentially originating from Amazon’s leadership. Notably, CEO Andy Jassy reportedly communicated these apprehensions to the Trump administration before the government’s crackdown (see: Reuters). Anthropic defended the security of its models, asserting that any reported “jailbreaking” methods could only expose simple vulnerabilities that most rudimentary models could detect. The firm insisted that Fable’s defensive capabilities were significantly more effective than those of any previously utilized model.
The sudden inaccessibility of these powerful AI tools has elicited a vehement reaction from policymakers in the European Union and the United Kingdom. Thomas Regnier, a spokesperson for the European Commission focusing on tech sovereignty issues, commented in an email statement that the incident exemplifies the need for Europe to bolster its technological independence. He emphasized the importance of existing EU legislation aimed at mitigating such dependencies, including the AI Act, the Cyber Resilience Act, and the NIS2 Directive—all vital tools for managing risks on European terms.
However, despite addressing cybersecurity risks, none of these regulations contain provisions explicitly designed to decrease reliance on foreign-made technology. Earlier this month, the European Commission introduced a long-awaited “tech sovereignty package,” which includes the Cloud and AI Development Act. This law seeks to effectively exclude American tech giants from public cloud contracts that involve the handling of sensitive data (see: EU Prepares Path for Shutting Out US Cloud Providers).
The discourse surrounding tech sovereignty has gained renewed vigor in Europe since the onset of the second Trump administration, fueled by escalating anxieties regarding American intentions toward strategic assets and fears that the U.S. government could implement a “kill switch” on the technologies extensively used across the globe. The closure of access to Anthropic models has been perceived as a tangible manifestation of this reality.
Benjamin Haddad, the French minister delegate for European affairs, articulated in a recent social media post that the actions taken by the Trump administration have accelerated a “geopolitical battle over AI.” He characterized the U.S. approach to AI as a significant national security instrument and a tool for global dominance, which ought not to be widely distributed. Haddad pointed out that critics of the EU’s push for tech sovereignty must confront the truth about the precariousness of relying on foreign technology.
Jordan Bardella, leader of France’s far-right National Rally party and a potential future presidential candidate, echoed this sentiment, noting that the unexpected decision underscored the criticality of AI as a matter of national sovereignty. He urged France to expedite its support for Mistral AI, the EU’s only legitimate competitor to Anthropic and OpenAI, in order to diminish dependence on external powers.
Conversely, perspectives from across the English Channel shared similar concerns, although the U.K. currently lacks a prominent AI alternative to put forth. Kanishka Narayan, the U.K.’s minister for AI and online safety, lamented in a video statement on social media that the restriction severed access to the globe’s most advanced AI model, a decision dictated by another nation rather than British authorities. Narayan underscored that the nation must begin treating issues of AI sovereignty with the same seriousness as other national security concerns, as the future of AI capabilities remains a defining challenge for Britain’s security strategy.
Narayan acknowledged the complexity of enhancing the U.K.’s AI capabilities, asserting there is no simple solution. He stated that Britain must elevate its AI capabilities, positioning this as a paramount political concern that requires clear understanding before external forces dictate the direction of its technological landscape.
Early access to Mythos was granted to the U.K.’s AI Safety Institute shortly after its launch, and the institute confirmed that it indeed represented a significant advancement in AI cybersecurity capabilities, similar to those demonstrated by OpenAI’s GPT-5.5-Cyber model. Meanwhile, it took nearly two months for Anthropic to agree to allow EU organizations access to similar models, an apparent reluctance on their part that has yet to be clarified. Prior to the recent U.S. directive, negotiations were underway, as confirmed by ENISA, the EU’s cybersecurity agency. Anthropic communicated its compliance with the export restrictions, and discussions with partners in EU nations continue to evolve.
The circumstances surrounding these U.S. export controls remain ambiguous. Anthropic’s contentious relationship with the administration likely influenced the government’s actions, particularly given the company’s hesitance to allow the deployment of Mythos in the context of fully autonomous weaponry or mass surveillance initiatives. Former U.S. AI czar David Sacks remarked that the export controls had no direct correlation to this contention, asserting that the actions stemmed from Anthropic’s reluctance to comply with safety requisites regarding the jailbreak risks.
President Trump has expressed a desire for the U.S. government to acquire stakes in leading AI companies amid ongoing developments in the sector. Both Anthropic and OpenAI are preparing for public offerings, following a lucrative IPO by SpaceX, the parent organization of xAI, which is another prominent player in the realm of AI.
The Financial Times recently reported that the U.S. National Security Agency had utilized its access to Mythos for executing cyber operations, further complicating the international landscape surrounding AI technologies and their implications on security and sovereignty.