The US government has taken an unprecedented step by ordering Anthropic, a leading AI company, to immediately block access to its two newest models — Claude Fable 5 and Claude Mythos 5. The decision, made public on June 12, 2026, cites concerns over national security, specifically the models’ potential to be used for cyberattacks and critical infrastructure breaches.
The directive, issued under export control regulations, prohibits any access by non-US citizens both abroad and within the United States, including foreign employees of Anthropic itself. As a result, the company was forced to shut down the systems entirely for all users, just three days after their highly anticipated debut on June 9. The models were hailed as breakthroughs in AI capability, but their rapid removal has sent shockwaves through the tech industry.
According to a report from Rzeczpospolita, which broke the story, the core of the government’s concern lies in the models’ advanced code analysis abilities, automated vulnerability detection, and capacity to execute complex tasks without constant human oversight. During internal lab tests in April, the Mythos architecture particularly demonstrated exceptional efficiency in finding and exploiting vulnerabilities in operating systems.
A ban that lasted three days
The speed and severity of the ban took even industry veterans by surprise. Anthropic had reportedly described the preview version of the model as „too powerful for public release” according to the company’s own internal documents, but claimed that commercial versions included robust safety filters. However, the US Department of Commerce received a report detailing a method for jailbreaking the Fable 5 model, potentially enabling it to analyze large codebases and pinpoint critical vulnerabilities.
From Washington’s perspective, such capabilities could be weaponized by hacker groups or state-backed cyber operations targeting sectors like energy, finance, or telecommunications. The administration’s move was described as a preemptive measure to prevent a potential catastrophe, even at the cost of stifling innovation.
Anthropic fights back
Anthropic has strongly rejected the allegations. The company insists that before the launch, both models underwent thousands of hours of red-teaming tests with institutions such as the UK’s AI Safety Institute and various US government agencies. According to Anthropic, the safety measures in Fable 5 are among the most effective in the industry, and the vulnerabilities highlighted by officials are known and reproducible in competing models as well.
– The administration made a drastic decision based on a single, narrow presentation that did not reflect the real level of risk – argued representatives of Anthropic, as quoted by Rzeczpospolita. The company believes the restriction was overly harsh and points out that the Pentagon previously labeled Anthropic a „threat to the supply chain,” though that earlier sanction was blocked by a court.
The new export decision, however, proved far more effective, effectively paralyzing the commercial use of Anthropic’s flagship models. Experts now worry that such restrictive and opaque actions could set a dangerous precedent for the entire AI industry.
Źródło: WNP.PL, Fot. Shutterstock AI Generator / Shutterstock
