Anthropic says it is off two AI prototypes launched earlier this week, Claude 5 Stories and 5 Storiesto comply with an export control order it received Friday afternoon from the US government citing national security concerns.
The unprecedented event is the latest landmark among Anthropic and the Trump administration. Although the company says the order asked it to suspend access to “any foreign national, whether inside or outside the United States, including Anthropic’s foreign employees,” it has removed access to all of its customers to ensure compliance.
Earlier this year, the Trump Department of Defense called Anthropic “supply chain risk” after Claude’s maker wanted to draw red lines about how the US military could use its technology. The designation effectively prevented government agencies and contractors from using Anthropic’s technology. Anthropic responded with open a case against the Trump administration.
On Tuesday, Anthropic went public The story of Claude 5a version of the company’s Mythos AI design with safeguards that prevent it from answering questions about cyber security, biology and chemistry. Before the public release, which Anthropic said it did in collaboration with the US government, the Mythos Preview AI example had a limited print run in April. The goal was to give companies and organizations the opportunity to use it strong network security capabilities to improve their protectionand raised concerns that the technology could be used by bad actors to create powerful hacking tools.
In a blog post on Friday, Anthropic says it received a letter from the US government at 5:21 pm ET. “The letter did not specifically address his national security concerns,” Anthropic wrote.
“Our understanding is that the government believes it has knowledge of a method to bypass, or ‘jailbreak’ Story 5,” the company added. “We reviewed a demonstration of this particular technique used to identify a small number of previously known vulnerabilities, small vulnerabilities. All of these vulnerabilities appear to be relatively simple, and we have found that other publicly available models can detect them as well without requiring a workaround.”
In a blog post, the company argued that it has implemented strong protections to reduce the potential for misuse of Claude Fable 5. Anthropic also claimed that the jailbreak the US government found for Claude Fable 5 was narrow, and would not make an attacker any more dangerous than they would have been with another AI model.
“So far, the government has only given us verbal evidence of the possibility of a small, non-universal jailbreak, which basically consists of asking the model to read specific rules and fix any software flaws,” the company said in its blog post. “Our understanding is that one incident of imprisonment was shared by the government.”
Spokesmen for the White House and the US Commerce Department did not immediately respond to WIRED’s request for comment.
Anthropic CEO Dario Amodei said in a policy essay earlier this week that he and the company support a fair, structured and transparent government process that will prevent the release of unsafe AI prototypes. In the company’s blog post on Friday, Anthropic said that “this action does not comply with those regulations.”
