Anthropic has disabled access to its newest AI models after the Trump administration ordered the company to halt their export, citing national security risks and possible cyber misuse.

The dispute centers on Claude Mythos 5 and Claude Fable 5, which Anthropic says were its most advanced systems to date. The models were launched only days before U.S. commerce secretary Howard Lutnick sent a letter on 12 June instructing the company to suspend their export and block their use by foreign nationals, including people living in the United States and overseas, according to Reuters, which saw a copy of the letter.

Anthropic said the only way it could comply with the export restrictions was to shut the models down entirely. The company said the government was worried the systems could be used by intelligence services in countries such as China and Russia, or adapted to help carry out cyberattacks.

The company also said officials pointed to a method of bypassing a safety measure in Fable that was meant to stop the model from being used to identify software vulnerabilities for malicious purposes. Anthropic said it did not agree that the bypass justified the action, arguing that the issue revealed only minor flaws that other widely available AI models can also detect.

In a statement, Anthropic said it supports government authority to block unsafe deployments, but only through a process that is transparent, fair and rooted in technical evidence. It said the current decision did not meet those standards and warned that applying such restrictions broadly could effectively stop the release of new frontier AI models.

The company said staff are now meeting with Trump administration representatives in an effort to find a path forward.

The move comes alongside a wider White House push to review frontier AI systems for security concerns before public release. Trump signed an executive order on 2 June that asks AI developers to voluntarily give the federal government access to certain advanced models for up to 30 days before they are shared more widely. The administration said the goal is to assess cybersecurity risks while keeping the United States competitive in AI.

The order says advanced AI brings national security considerations that require coordination across federal departments and agencies. It also says the administration wants to promote cybersecurity innovation while maintaining U.S. leadership in artificial intelligence.

David Sacks, a venture capitalist and former White House adviser, said on X that the review process would focus on models that represent a meaningful jump in cyber capability, not small version updates. He said the shorter 30-day review window, down from a previously discussed 90-day period, would let companies follow the voluntary framework without slowing releases.

The executive order also calls for the creation of an AI cybersecurity clearinghouse, working with industry and critical infrastructure operators. The clearinghouse would help scan for software vulnerabilities and distribute patches when problems are found.

Other federal agencies are expected to issue directives and guidance that would speed up cyber defense efforts, expand defensive AI tools and improve access to cybersecurity services for government agencies and critical infrastructure providers. The order also directs agencies to look for grant funding that could support advanced vulnerability-detection research and to expand hiring routes for cybersecurity specialists in the federal government.

The Anthropic case is likely to become an early test of how aggressively the U.S. government will use its new authority, and how much latitude AI developers will have as frontier models become more capable and more closely tied to national security policy.