US Orders Anthropic to Disable Frontier AI Models for All Foreign Nationals

In an unprecedented escalation of the global technology race, the United States government has issued an emergency export control directive forcing AI research lab Anthropic to disable access to its newest frontier models for all foreign nationals. The sweeping order, which targets the newly released Claude Fable 5 and Mythos 5 models, requires Anthropic to cut off access not only to users outside the United States but also to non-US citizens residing within the country. The directive marks the first time Washington has weaponized export controls to physically sever global access to a cloud-based Large Language Model, transforming commercial AI infrastructure into a tightly guarded national security asset overnight.[1][2]

The ban arrived a mere three days after Anthropic launched Fable 5 and Mythos 5 to widespread industry acclaim. Released on June 9, the models represented a massive leap in artificial intelligence capabilities, topping nearly every industry benchmark for software engineering, advanced reasoning, and long-context memory. Fable 5 was made available to the general public at an aggressive price point of $10 per million input tokens, while Mythos 5—a version with lifted safeguards—was restricted to a small cohort of approved cybersecurity and biology researchers. By June 12, the US government had intervened, citing severe national security risks and potential vulnerabilities that could be exploited by foreign adversaries.[3][5][6]

Anthropic complied with the 5:21 p.m. directive, taking the models offline globally and returning error messages to international developers who had already begun integrating the system into their applications. However, the San Francisco-based company publicly pushed back against the government's rationale. In a detailed statement, Anthropic argued that the vulnerabilities identified by federal authorities were minor, benign, and present in virtually every other frontier model on the market. The company maintained that if the government's new standard for 'jailbreaks' were applied evenly across the industry, it would effectively halt all new AI model deployments from every major American provider.[2][7]

The models themselves were designed with some of the strictest safety guardrails ever implemented in a commercial AI product. Fable 5 utilized advanced safety classifiers that would automatically block sensitive requests—particularly those related to offensive cybersecurity, chemical synthesis, or biological weapons—or route them to the older, less capable Opus 4.8 model. In pre-release testing, Fable 5 reportedly complied with zero harmful single-turn requests, even when subjected to dozens of sophisticated jailbreak techniques. Despite these built-in defenses and a newly implemented 30-day data retention policy designed to catch malicious actors, federal regulators determined the risk of the models falling into foreign hands was too great.[3][5][7]

The technical capabilities that triggered Washington's alarm are formidable. Mythos-class models demonstrated an unprecedented aptitude for offensive cyber operations, proving capable of finding zero-day exploits, navigating complex network architectures, and executing multi-step digital attacks with minimal human oversight. In one corporate test, Fable 5 successfully executed a codebase-wide migration of a 50-million-line Ruby environment in a single day. The models also outperformed specialized, peer-reviewed AI systems in scientific research tasks despite being a fraction of the size. For US intelligence and defense agencies, these capabilities crossed the threshold from commercial software to dual-use strategic weaponry.[3][5][6]

The sudden blackout has sent shockwaves through the global technology sector, exposing the profound fragility of building businesses on top of American-controlled API endpoints. The directive does not exempt the United States' closest intelligence allies; developers and enterprises in the United Kingdom, Canada, Australia, and the European Union found their access revoked without warning. For international startups that had spent the previous 72 hours migrating their tech stacks to Fable 5 to take advantage of its superior coding and reasoning abilities, the abrupt shutdown resulted in broken applications and a frantic scramble to revert to older systems.[1][4][7]

This geopolitical shock has immediately accelerated calls for 'AI sovereignty'—the movement urging nations and corporations to build and host their own artificial intelligence infrastructure rather than relying on Silicon Valley. Isaacus, an Australian-based legal AI research company, pointed to the Anthropic ban as validation of their strategy to rely exclusively on air-gapped, self-hostable models. 'In effect, any application depending on US-based LLMs is subject to being shut down at any moment as a result of an export control directive,' the company noted, emphasizing that mission-critical intelligence cannot be outsourced to platforms vulnerable to the whims of Washington's foreign policy.[4]

The long-term implications of the directive threaten to fracture the global AI ecosystem. By establishing that cloud-based frontier models are subject to immediate export bans, the US government has effectively signaled that non-American entities cannot safely rely on American AI for critical infrastructure. This reality is expected to drive massive international investment into open-weight models and sovereign supercomputing clusters in Europe, the Middle East, and Asia. While the US aims to maintain its strategic edge by hoarding its most capable systems, the ultimate result may be the rapid acceleration of competitor models designed specifically to operate outside the reach of American export controls.[1][4][6]