Anthropic Shuts Down Most Advanced Models
On June 9, 2026, Anthropic, one of the leading AI development companies, released its latest models, Claude Fable 5 and Mythos 5. Days later, the Trump administration directed Anthropic to restrict access to the models for foreign nationals, citing an unspecified risk to national security. Anthropic complied by withdrawing access to the targeted AI models for all users, arguing that doing so was the only feasible way to comply with the administration’s directive. This directive is the latest in a series of regulatory disagreements between Anthropic and the White House, and it introduces significant long-term implications for the governance of AI development.
The Dispute Over the Mythos Launch
Part of Anthropic’s highly capable “Mythos class” of AI, Fable 5 and Mythos 5 outperformed all other Anthropic AI models in company testing. To make this leap in capability safe for the general public, Anthropic built Claude Fable 5 by placing safeguards on responses to hazardous requests. These safeguards are designed to prevent malicious activities, such as cyberattacks, that exploit the model’s technical sophistication. Mythos 5 is the less restricted version of Fable 5, deployed exclusively to vetted corporations and government agencies.
The rollout of both AI models was cut short just three days after launch, however, when they were temporarily pulled offline for all users. According to an X post by White House Advisor David Sacks, a trusted partner of Anthropic and the U.S. government identified a breach in Fable 5’s safeguards. The government followed up by privately asking Anthropic to address the breach or recall the model altogether, a request that Anthropic’s CEO reportedly refused. Following negotiations with the administration, Anthropic restricted access to the models to comply with the directive. Anthropic later pushed back on the severity of the situation, arguing that other publicly available AI models possess the same capabilities that the reported breach revealed.
The situation stems from a broader effort by the Trump administration to keep advanced AI out of the hands of foreign competitors. Given Fable 5’s technical sophistication, the government grew concerned that a security flaw would grant foreign adversaries access to the model’s cyber capabilities. As a preventative measure to mitigate the risk, the administration instructed Anthropic to block access for all non-U.S. citizens.
What Is Driving Tensions Between Anthropic and the White House?
There have been long-running tensions between Anthropic and the Trump administration. In February, a Wall Street Journal report revealed how Claude, Anthropic’s AI tool, was instrumental in the U.S. military’s capture of former Venezuelan President Nicolás Maduro in January 2026. The operation breached Anthropic’s guidelines that forbid the usage of Claude to carry out violence, creating tensions around Anthropic’s $200 million contract with the military. Later that month, U.S. Secretary of Defense Pete Hegseth demanded that Anthropic alter usage policies that restricted military applications of its AI models. The AI company refused the demands, leading the government to terminate the contract and designate Anthropic a supply chain risk.
Following President Trump’s recent order to recall Fable 5 for non-citizens, Anthropic released a statement that outlined several points. First, the company claimed to have worked with the government for “thousands of hours in total” to ensure safeguard reliability prior to release. Second, they asserted that internal testing hadn’t found any significant safeguard breakthrough, and that the minor reported breaches could be reproduced by any other publicly accessible AI model. Third, Anthropic said that the government failed to cite a specific case to justify its directive. Some contend that these details suggest government targeting as the reasoning behind the action.
To counteract criticism of the White House, Sacks asserted that the directive isn’t tied to historical disagreements with Anthropic. He instead argued that Anthropic’s position was hypocritical, as it brands itself as an AI safety company while disregarding concerns of a safeguard breach. According to a senior White House official, the government intervention was a last resort after “begging them for hours to work with us.”
The Future of AI Regulation
As AI capabilities develop, the dangers of unrestricted access to AI become more pronounced. Anthropic’s Claude Mythos was identified as the first-ever restricted AI model due to its threatening cybersecurity potential, raising concerns about potential hazards to U.S. national security infrastructure. Broader concerns around AI development and cybersecurity recently prompted the creation of Project Glasswing, a coalition that convenes top corporations such as Amazon, Apple, Google, Microsoft, Nvidia, and JPMorganChase to preemptively identify vulnerabilities in global operating systems and web browsers.
The government order itself carries long-term implications. By leveraging an export control directive, an action that the White House can employ to protect national security interests, the Trump administration sets a key precedent. Anthropic’s restrictions mark the first time an export control directive has been used to directly control AI access for non-Americans, potentially making it harder for leading AI companies to acquire foreign talent due to regulatory instability. Internationally, the expansion of export controls has unnerved foreign allies, who worry that access to American AI could become subject to quickly changing political decisions.
As the dust settles on this unprecedented standoff, the advent of the Mythos class and its regulatory struggles may become a watershed moment carrying profound implications for AI development and international security.