Anthropic has suspended access to its newest language models, Claude Fable 5 and Mythos 5, following an instruction from U.S. national security authorities. The BBC reported the development, and Anthropic confirmed the suspension in a statement posted on its website.
According to the company, the government ordered that Fable 5 access be blocked for foreign users, and as a result Anthropic had to disable both Fable 5 and Mythos 5 for all of its customers with immediate effect. The company said U.S. national security agencies did not point to a specific threat, but identified a method that could bypass the models’ safety constraints.
Anthropic said it examined the technique the government presented and judged that it mainly exposed a few previously known, minor vulnerabilities. The company also stated that other publicly available models are capable of identifying similar weaknesses without actually circumventing safety controls.
Controversy and test results
Fable 5 was already the subject of debate before its public release. Anthropic provided limited early access to several organisations in April, and the company itself warned the model was too powerful for general release because of its potential to break into computer systems and exploit security flaws. Critics argued that this claim was exaggerated marketing.
Tests by a UK government MI security institute indicated the model was able to evade defensive systems in 73 percent of cases. Gina Neff, a professor of responsible AI at Queen Mary University of London, said this represented a "step change" in cyber capabilities.
The European Union, which received access to Mythos in June, has also weighed in, stressing the importance of Europe’s technological sovereignty. A European Commission spokesperson said the situation is under assessment while the Commission has recently announced measures to reduce technological dependence on U.S. and Asian providers.
Strained relationship with the U.S. government
Relations between Anthropic and the Trump administration have been tense. Donald Trump publicly criticised the company, and Pete Hegseth, a former defence official, labelled Anthropic an "supply-chain risk" — a designation normally reserved for firms from rival or hostile countries and not previously applied to an American company.
Anthropic sued the Pentagon over that designation. A U.S. judge previously ruled the Pentagon’s directive was not enforceable, allowing government agencies to continue using the company’s tools during the litigation; however, that ruling did not prevent the current national-security-based access restrictions.
Why this matters
The dispute around Fable 5 and Mythos 5 sits at the intersection of artificial intelligence and cybersecurity: regulatory action, concerns about sovereignty, and legal and security conflicts between technology firms and state institutions are shaping which tools are available and how they can be used globally. The case will remain important to regulators and industry players monitoring the governance of powerful AI models.
Tags: European Union, Donald Trump, cybersecurity, artificial intelligence, lawsuit, Pentagon, national security, sovereignty, language model, Pete Hegseth
