Anthropic AI restrictions reignite debate over AI sovereignty

US government restriction on foreign access to Anthropic’s Fable 5 and Mythos 5 models has triggered broader concerns about AI sovereignty among American allies. The move has raised questions about whether governments and companies outside the United States can reliably depend on frontier AI systems controlled by US firms and subject to national security restrictions.

The directive reportedly required Anthropic to prevent non-American users, including foreign nationals working inside the company, from accessing the models. Anthropic responded by suspending access more broadly, stating that this was the only practical way to comply with the directive.

The immediate dispute centres on concerns that Fable 5 could be jailbroken and used beyond its intended safeguards. However, the broader impact extends beyond one company or one model. Governments, security agencies and companies that had secured access to Anthropic’s most advanced systems reportedly saw those permissions withdrawn overnight.

The Anthropic cutoff has been particularly sensitive for US allies. Reports indicate that the restrictions extended even to partners in the Five Eyes intelligence alliance, including Australia, the UK, Canada and New Zealand. The UK’s AI Security Institute, which has played a leading role in testing and evaluating advanced models, was also reportedly affected.

The episode has strengthened arguments that countries may need greater sovereign AI capabilities rather than relying heavily on frontier models controlled by foreign providers. For allies, the question is not only whether they can access advanced AI systems today, but whether that access can be withdrawn suddenly because of US policy decisions, export controls or national security interventions.

The episode also highlights a difficult policy trade-off for the United States. The United States has a strategic lead in frontier AI and may seek to prevent the most capable systems from being misused or accessed by adversaries. Yet applying broad restrictions to allies and foreign employees could damage trust, disrupt research and push other countries to accelerate domestic AI development.

For middle powers, building AI sovereignty will not be straightforward. Training frontier models requires advanced chips, large-scale compute infrastructure, talent and capital, all of which remain concentrated in a small number of countries and firms. Restrictions on chip exports could also limit the ability of allies to build independent alternatives.

The dispute, therefore, points to a wider geopolitical shift in AI governance. As frontier AI models become more capable, access to them is increasingly being treated as a matter of national security. That could force governments to rethink procurement, cloud dependence, AI testing partnerships and long-term strategies for technological sovereignty.

Why does it matter?

The episode illustrates how access to advanced AI systems is becoming a strategic issue rather than simply a commercial service. As frontier models become increasingly important for research, cybersecurity, defence, innovation and economic competitiveness, governments are beginning to view access controls through the lens of national security and geopolitical influence.

The case also highlights a growing tension between AI leadership and international trust. While countries may seek to restrict access to powerful systems to prevent misuse, abrupt limitations affecting allies can encourage efforts to build domestic AI capabilities and reduce dependence on foreign providers. As a result, debates about AI sovereignty, technological autonomy and strategic resilience are likely to become increasingly central to digital policy worldwide.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!