The U.S. Department of Commerce’s recent restrictions on access to AI models like Anthropic’s Claude 2 and Claude 3 have sparked a debate over how to regulate advanced artificial intelligence, according to a June 12 directive issued to the company. The order, which requires Anthropic to block foreign nationals from using these models, marks a shift in how governments define “export” in the digital age.
What triggered the conflict between Anthropic and the U.S. Commerce Department?
Anthropic, a U.S.-based AI company, disclosed on June 12 that the Department of Commerce had instructed it to halt access to its Claude 2 and Claude 3 models for all foreign nationals, including employees with non-U.S. citizenship. The company stated this requirement stems from a broader executive order designating AI models as potential supply chain risks. While the Commerce Department did not explicitly mention Anthropic in its public statements, legal experts note the directive aligns with a 2023 executive order targeting “advanced AI systems” deemed critical to national security.
According to a statement from Anthropic, the restriction applies to “any foreign person,” including employees, customers, and partners. The company emphasized that “access to other Anthropic models remains unaffected,” but the move has raised concerns about the broad implications for global AI collaboration.
How do the new rules redefine “export” in the AI era?
The Commerce Department’s guidance expands the definition of “export” to include not just the transfer of source code but also access to AI models themselves. Under the agency’s rules, providing a model to a foreign national—regardless of where the model is hosted—qualifies as a “deemed export,” requiring explicit approval. This approach reflects a growing trend among governments to treat AI capabilities as strategic assets, rather than just technical tools.
Mark Lashoda, a former federal prosecutor and AI policy consultant, explained that the rules “shift the focus from physical data movement to control over AI capabilities.” This evolution has significant implications for companies offering AI as a service (AaaS), as it complicates compliance for businesses with international teams and customers.
What challenges do companies face under the new regulations?
The restrictions pose immediate operational challenges for firms like Anthropic. A significant portion of its workforce holds non-U.S. citizenship, and some employees have direct access to the underlying code of its models. More fundamentally, the rules create a paradox: while AI models are often accessed via APIs, verifying a user’s nationality in real time is nearly impossible.
Ulyana Golyunov, CIO of consulting firm Acceligence, noted that “there’s no reliable way to confirm a user’s nationality through API calls. Additionally, 75% of U.S. citizens don’t even hold a passport.” This technical limitation forces companies to adopt a cautious approach, potentially blocking access for all users to avoid legal risk.
Why is this shift significant for global AI competition?
The regulations reflect a broader geopolitical strategy to control the spread of cutting-edge AI technologies. By treating access to models as a form of export, the U.S. aims to limit adversaries’ ability to develop their own AI systems. However, experts warn this could accelerate the development of alternative AI ecosystems.
Valence Howden, an advisor at Info-Tech Research Group, argued that the rules “signal a fundamental shift in how nations view AI. It’s no longer about controlling data or code, but about controlling who can harness the power of AI.” This dynamic is already influencing corporate strategies, with some companies exploring partnerships with non-U.S. AI providers to mitigate risks.
What are the long-term implications for AI governance?
The conflict highlights the tension between innovation and national security in the AI era. While the U.S. aims to protect its technological edge, the regulations risk stifling global collaboration and creating a fragmented AI landscape.
Sanjit Bhagat, a senior analyst at Greyhound Research, emphasized that “the rules are built for a world where technology is tied to physical assets, but AI operates in a domain where value lies in capabilities, not just code.” As governments grapple with these challenges, the balance between security and open innovation will define the future of AI governance.
For now, companies must navigate a rapidly evolving regulatory environment. Anthropic’s decision to suspend access for all users underscores the uncertainty surrounding these rules, as businesses adapt to a new reality where AI access can be restricted at any time, regardless of location or intent.