The United States Restricts Foreign Access to AI Models for the First Time and the Challenge of Technological Sovereignty

Anthropic is currently one of the largest AI companies in the United States and is widely regarded as a direct competitor to OpenAI. The company's flagship product is Claude, an AI assistant capable of engaging in conversations, analyzing documents, supporting research, generating content, and programming. Claude is currently deployed extensively for individual users, businesses, and software developers through Anthropic’s platform as well as the cloud services of Amazon and Google. In early June, Anthropic introduced Claude Fable 5 and Mythos 5, the company's most advanced generation of AI models, belonging to the capability tier that Anthropic refers to as the “Mythos-class.” According to Anthropic, Fable 5 is the most powerful model the company has ever released for broad use, delivering outstanding performance in evaluations involving programming, scientific research, image processing, and tasks requiring complex analytical capabilities. The company stated that Fable 5 capabilities are particularly notable in tasks involving extended multi-step processes.

However, only a few days after Fable 5 was announced, Anthropic stated that it had received an export control directive from the U.S. Government requiring the suspension of access to Fable 5 and Mythos 5 for foreign nationals. According to the company, the U.S. authorities did not specify their national security concerns, but Anthropic understood that the government was concerned about the existence of a “jailbreak” method (a technique for bypassing the safety safeguards built into an AI model) that could allow users to use Fable 5 to identify software vulnerabilities. These concerns stem from the fact that Fable 5 is considered one of the most advanced AI models currently available and is being deployed to hundreds of millions of users. Some experts argue that AI models belonging to the Mythos class, if misused, could significantly accelerate the process of vulnerability discovery or support sophisticated cyberattack activities.

Anthropic stated that, prior to the release of Fable 5, the company had worked with the U.S. Government and various other organizations during the safety evaluation process, while also implementing safeguards designed to restrict the use of the model in sensitive areas such as cybersecurity. The company further asserted that it had been provided only with “verbal evidence” regarding a limited potential vulnerability. Anthropic argued that this did not constitute a sufficient basis for withdrawing a commercial model that was already being widely deployed. Nevertheless, the company stated that it was still required to disable Fable 5 and Mythos 5 in order to ensure compliance with the directive issued by the U.S. Government.

The action taken against Anthropic is widely regarded as a notable turning point in U.S. technology control policy. For many years, Washington’s export control measures have primarily focused on semiconductors, semiconductor manufacturing equipment, and tools used for AI development. The objective of these measures has been to restrict access to foundational technologies that could be used to develop advanced AI systems. However, the directive issued to Anthropic suggests that the scope of control is expanding from hardware to AI models themselves. For years, the United States concentrated on controlling AI-related chips and tools rather than restricting foreign users’ access to AI. The requirement that Anthropic suspend the provision of Fable 5 and Mythos 5 to foreign nationals on national security grounds is regarded as one of the first indications that Washington has begun to treat advanced AI models themselves as objects of regulatory control. This shift reflects a new U.S. perspective on AI. Whereas technological capability was previously assessed primarily through the ability to manufacture chips or possess computing infrastructure, advanced AI models are now increasingly being viewed as strategic assets capable of generating economic, technological, and national security advantages. In this context, access to AI may gradually become a policy issue of comparable importance to access to semiconductors and other core technologies.

The U.S. Government’s first-ever restriction on access to a commercial AI model has also sparked considerable debate regarding the impact of this policy on the global AI market.

From the first perspective, supporters argue that restricting foreign nationals’ access to advanced AI models is necessary to protect U.S. national security in a context where AI systems are increasingly capable of assisting in vulnerability discovery, source code analysis, and the performance of complex cybersecurity tasks. Under this approach, controlling access to advanced AI models may help reduce the risk of technology being exploited for harmful purposes, while also protecting the technological advantages and competitive capabilities of U.S. companies. In the short term, such measures may also contribute to maintaining the United States’ leading position in the global AI race.

However, others argue that restrictions on access to technology may produce unintended consequences. If businesses, research institutions, and users outside the United States no longer have stable access to advanced AI models, they may have stronger incentives to seek alternative solutions or develop domestic AI systems in order to reduce their dependence on U.S. technology. In the long term, this trend may encourage the emergence of separate AI ecosystems in different regions of the world, increasing technological fragmentation and reducing the interconnectedness of the global AI market. At the same time, U.S. AI companies may risk losing a portion of their international market share if customers shift to alternative providers.

It is evident that AI is gradually moving beyond its role as a mere tool for work support or innovation to become a technology of strategic significance. Similar to semiconductors, energy, and data, the ability to control access to advanced AI models is increasingly being viewed by states as a matter related to national security, competitiveness, and geopolitical influence.

In response to these developments, the European Commission (EC) stated that it is assessing the practical implications of the export control directive imposed by the United States on Anthropic. In a statement issued on 14 June, EC spokesperson Thomas Regnier noted that next-generation AI models offer numerous benefits, including applications for cyber defense, while also giving rise to serious cybersecurity concerns that need to be addressed.

However, Mr. Regnier emphasized that responses in this area should not be discriminatory toward U.S. partners. According to the EC, a country's unilateral restriction of access to strategic technologies may have significant implications for users and businesses in other countries that depend on cross-border technology services.

Notably, Mr. Regnier argued that the case provides further evidence of the need for Europe to strengthen its technological sovereignty. According to the EC, the U.S. decision regarding Anthropic demonstrates that access to core digital technologies is increasingly subject to the national security considerations and foreign policy priorities of the country that controls such technologies. In this context, building independent technological capabilities is becoming an increasingly important priority for the European Union.