US Orders Anthropic to Block 2 AI Models from Foreign Nationals, Citing Security

Legal and compliance teams watching this case will immediately question the statutory basis for a blanket ban with no public record of detailed justification. The directive applies extraterritorially, affecting foreign individuals on US soil, and appears to sidestep established export control frameworks—exposing the government to potential lawsuits and sparking debates about the boundaries of national security power in the AI era.

On June 12, 2026, the United States government issued an unprecedented order to AI company Anthropic, compelling it to immediately block access to its latest AI models for all foreign nationals worldwide. The order, received at 5:21pm Eastern Time, targeted the publicly released Fable 5 model and the non-public Mythos 5, citing national security concerns but providing no detailed justification. The ban extends to foreign individuals physically present in the US, including foreign employees at Anthropic itself, forcing the company to instantaneously revoke access on a global scale. This sudden directive marks the most aggressive use of executive authority to restrict AI access in US history, raising profound questions about the balance between innovation, security, and international collaboration.

“

The order, received at 5:21pm Eastern Time, targeted the publicly released Fable 5 model and the non-public Mythos 5, citing national security concerns but providing no detailed justification.

At the heart of the order lies the unique capability of Anthropic's Mythos AI architecture, which has demonstrated a remarkable proficiency in detecting software vulnerabilities, some of which have lain dormant in critical systems for decades. This capability has been leveraged by US authorities and select corporate partners to harden infrastructure against cyber threats. However, from the outset, experts warned that such a powerful tool could become a devastating cyberweapon if it fell into the wrong hands. Fable 5, released just days earlier, is based on Mythos technology but with its cybersecurity and biotechnology capabilities deliberately blocked. Mythos 5, the full uncensored version, remains strictly under US government and partner control. The order effectively prevents any non-US person from exploiting Fable 5’s underlying code review capabilities, even if those abilities are intentionally limited.

Anthropic’s response, published in a blog post, reveals a company grappling with incomplete information. The firm disclosed that it had received only partial instructions and had reviewed a classified report that likely triggered the order. Its own assessment concluded that the report referenced only a limited ability of the AI to review specific program code and correct errors—a capability it argues is also present in rival models like OpenAI’s GPT-5.5. Anthropic explicitly disagreed with the classification of this capability as a justification for a blanket foreign ban, hinting at a potential legal challenge or an industry-wide precedent. The company’s statement stressed that it had to comply at short notice, underscoring the sweeping and unpredictable nature of the government’s intervention.

The implications ripple across the technology sector, defense community, and international relations. For cybersecurity professionals, this move signals that advanced AI tools for vulnerability detection are now firmly categorized as dual-use technologies, subject to export-control-like restrictions. It disrupts global collaborative efforts to patch critical flaws, as foreign security researchers and enterprises are abruptly cut off from leading-edge AI assistance. The decision may also accelerate a fragmentation of the AI landscape, pushing other nations to develop or restrict their own sovereign models, mirroring the geopolitical dynamics seen in the semiconductor industry.

Legally, the order’s basis remains opaque, inviting scrutiny under domestic and international law. It appears to bypass established processes for technology export controls, such as the International Traffic in Arms Regulations (ITAR) or the Export Administration Regulations (EAR), by imposing an instant ban without advance notice or public consultation. The extraterritorial application—covering foreign nationals within US borders—could face constitutional challenges related to due process and equal protection. For the AI industry, the episode sets a chilling precedent: even publicly released, supposedly safe models can be retroactively pulled from the global market at a moment’s notice, potentially undermining investor confidence and commercial deployment strategies.