Bottom line: The US government blocks two high-performance Anthropic AI models for foreign nationals over concerns about a workaround to security restrictions — a step Anthropic criticizes as non-transparent and technically unjustified.
The US government has instructed Anthropic to block access to the AI models Fable 5 and Mythos 5 for all foreign nationals — including persons residing in the country and employed at Anthropic. The reason is concern about a method with which security restrictions in the software could be circumvented.
Following a US government directive, Anthropic has blocked foreign nationals’ access to its recently released flagship models Fable 5 and Mythos 5. The directive was issued with reference to national security. The company subsequently had to restrict access for all users domestically and internationally to ensure compliance — including employees with foreign citizenship.
Mythos 5 is a non-public model with advanced cybersecurity capabilities: it can identify long-undetected software vulnerabilities and has so far been deployed by US agencies and select corporate partners for system hardening. Fable 5, the newly released public version, is based on this Mythos technology but explicitly blocks cybersecurity and biotechnology capabilities. The government’s concern is that such capabilities in the wrong hands could be misused as a cyberweapon.
Anthropic stated that it received only partial information from the government. The company analyzed a report that likely triggered the directive and concluded that it concerns a limited capability to have the AI review specific program code and correct errors — a capability that also exists in models from other providers such as OpenAI’s GPT-5.5. Anthropic argues that the security safeguards in Fable 5 have been extensively tested and the blocking of software for hundreds of millions of users is not justified.
The case raises questions about regulatory consistency: Anthropic CEO Dario Amodei recently advocated himself for governmental blocks on potentially dangerous AI software — but demanded transparent and technically fact-based procedures. This, he argues, is not present in the current situation. In parallel, Anthropic already entered into conflict with the US government by refusing to deploy its models in autonomous weapon systems and for mass surveillance in the US. The Pentagon subsequently classified the company as a supply chain risk, restricting use of Anthropic software by government agencies. Anthropic is suing against this classification.
Source: www.it-daily.net · Published June 13, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.1.