Impact of the US Department of Commerce Directive on Anthropic AI Models
Recent U.S. Department of Commerce Directive targeting Anthropic’s advanced artificial intelligence (AI) systems highlights how trade regulations are being applied to global cloud-based technologies. This directive was issued because the US government learned of a way to bypass Clause Fable 5, with no further details on national security concerns. The US government and Anthropic have a disputed recent past. This directive led Anthropic to stop access worldwide to two of its models- Claude Fable 5 and Claude Mythos 5. Because compliance required preventing access by foreign persons but there is no way to separate users by citizenship, Anthropic had to withdraw the models entirely to ensure compliance.
The U.S. Commerce Department’s Export Administration Regulations include a “deemed export” rule, which treats the provision of controlled technology to a “foreign person” (anyone who is not a U.S. citizen or green-card holder) as an export to that person’s country of nationality. This rule extends the concept of an export beyond physical borders, treating access by individuals as equivalent to transferring technology to their country of nationality, even when the technology remains hosted entirely within a country, in this case the United States. This internalises the border within digital systems and shifts regulatory focus from geography to individuals.
This creates significant enforcement challenges for AI services delivered online. AI providers cannot reliably verify the citizenship status of users in real time, and location data alone cannot determine nationality. Therefore, it is difficult to selectively restrict access based on citizenship alone.
This has broader implications for AI governance. There is a gap between policy interpretations of exports and the technical realities of cloud-based services, which have imperfect identity verification and are globally accessible. It also creates uncertainty for users and organisations that rely on stable access to AI systems for research, analysis, and operational use, as policy changes in one jurisdiction can have immediate global effects.
F
or Australia, the incident highlights the risks of dependence on foreign-owned AI systems and infrastructure. Australian universities, government agencies, businesses, healthcare providers, and researchers increasingly rely on advanced AI models hosted and controlled overseas. The withdrawal of access demonstrates that even close allies can be affected by regulatory decisions made in another jurisdiction, with little ability to influence those outcomes. This creates risks for continuity of operations, research capability, innovation, and economic competitiveness.
The incident also strengthens the case for greater AI sovereignty. Sovereign AI does not necessarily require Australia to develop frontier models entirely from scratch, but it does require reducing dependence on a small number of foreign providers. This includes investing in domestic data infrastructure, computing capacity, secure cloud services, open-source AI capabilities, and the workforce needed to develop, deploy, and govern these technologies. Building sovereign capability would improve Australia's resilience to external policy changes and help ensure continued access to critical AI tools and services.
More broadly, the issue highlights the tension between national security objectives and the operational characteristics of modern AI infrastructure. The challenge for policymakers is to design export-control frameworks that remain enforceable without producing unintended global disruption, while also encouraging diversified and resilient AI ecosystems. Globally, the key policy question is how to balance the benefits of access to leading international AI systems with the need for greater technological sovereignty and long-term strategic resilience.
Questions:
How should rights and access be governed in a world of borderless AI?
If AI services operate globally but are controlled by the laws of a single country, should access be determined by territorial laws, national sovereignty, or broader principles of digital and human rights?
Is access to advanced AI a fundamental human right, or should it be one, and what should its scope be?
As AI becomes essential for research, education, healthcare, creativity, and economic participation, should individuals and organisations have a protected right to access these technologies, or can governments legitimately restrict access for national security reasons?
Should AI regulation focus on misuse of technology rather than innovators and broadly users of a technology?
Do export controls based on nationality unfairly presume risk without evidence of wrongdoing, creating a form of “no-fault” liability for users and companies? What level of proof should be required before access is restricted?
What rights do users have alongside AI innovators and platform providers?
When users create research, businesses, or creative works using AI systems, what rights and protections should they have if access is withdrawn? How should user interests be balanced against intellectual property rights, commercial interests, and regulatory obligations?
What are the trade-offs between national security and open access to AI?
While export controls may help prevent the spread of strategically sensitive technologies, they can also disrupt innovation, research, economic activity, and international collaboration. How should policymakers balance these competing objectives, and what role should sovereign AI capability play in reducing these risks?
Sources:
Justin Downes. The AI That Scared the US Government Into Hitting the Kill Switch. 18 June 2026. https://medium.com/@justin.edgewoods/https-www-softpagecms-com-category-technology-8676b6995340
Congressional Research Service. Pentagon-Anthropic Dispute over Autonomous Weapon Systems: Potential Issues for Congress. 13 March 2026. https://www.congress.gov/crs_external_products/IN/PDF/IN12669/IN12669.1.pdf
Order of the United States District Court, Northern District of California in Anthropic PBC v. US Department of War et al. 26 March 2026. https://assets.bwbx.io/documents/users/iqjWHBFdfxIU/rYKRX7EU4j5U/v0
Announcement by Anthropic. 12 June 2026. Statement on the US government directive to suspend access to Fable 5 and Mythos 5. https://www.anthropic.com/news/fable-mythos-access
Jon Markman. How The Anthropic Crackdown Backfired Into An AI Boom. Forbes. 17 June 2026. https://www.forbes.com/sites/jonmarkman/2026/06/17/how-the-anthropic-crackdown-backfired-into-an-ai-boom/
Gilad Abiri and Dimitry Vladimirovich Kochenov. Banned From the Future: Keys to AI Technology and the Future of European Citizenship. Verfassungsblog: On Matters Constitutional.15 June 2026. https://verfassungsblog.de/anthropic-mythos-ban-foreigners-citizenship/
Ari Hawkins. Trump’s Anthropic restrictions may be illegal. Politico. 18 June 2026. https://www.politico.com/news/2026/06/18/trump-anthropic-ai-export-controls-00966118
Bernadette Hyland-Wood, Dimitri Perrin, and Michael Guihot. The US government can shut off access to AI at will. What does this mean for Australia? The Conversation, Australia edition. 17 June 2026. https://theconversation.com/the-us-government-can-shut-off-access-to-ai-at-will-what-does-this-mean-for-australia-285480
