White House-ordered AI export controls spark debate over regulation and feasibility
6 verified4 unconfirmed2 contested
The Trump administration has imposed export control restrictions on Anthropic’s new Claude Fable 5 and Mythos 5 AI models, forcing the company to take them offline. Officials cited national security concerns, alleging that Fable 5’s safeguards can be bypassed to access dangerous capabilities in chemistry, biology, and cybersecurity. The National Security Agency reportedly concluded that such jailbreaks are possible. Anthropic has pushed back, arguing the risks are minimal and that the effects of any jailbreaks are limited. The company held a technical meeting with the Commerce Department and the Office of the National Cyber Director but has not secured an agreement to restore the models. Experts say the situation is unprecedented, as this appears to be the first time US export controls have been applied to restrict access to an AI model in this manner. Many observers note that similar capabilities may soon exist in models from other companies, making the broader governance questions unavoidable.
What’s verified
The White House placed an export control restriction on Anthropic’s Mythos 5 and Claude Fable 5 models, forcing the company to block access for all foreign nationals and effectively take both models offline.
The administration believes Fable 5’s guardrails can be disabled to enable access to Mythos 5’s advanced capabilities related to cybersecurity, chemistry, and biology.
The National Security Agency concluded there are ways to bypass Fable 5’s safeguards.
Anthropic contends the jailbreak risks are minimal and has argued that the administration’s concerns are overblown.
Anthropic held a technical meeting with the Commerce Department and the Office of the National Cyber Director to discuss the issue.
Experts say this is the first known instance of US export controls being used to control access to an AI model in this way, and that the existing regulatory framework is ill-suited to such cases.
Where accounts differ
Disagreement exists over how much warning Anthropic received before being ordered to take down its models. One account says Anthropic was given 90 minutes to comply, while another claims the administration implored the company “for hours.”
The reason for the White House action varies by account: one source points to a jailbreak discovered by Amazon, another claims the same results could be achieved with OpenAI’s ChatGPT 5.5, a third says a China-linked group may have accessed Mythos (though no confirmation of a jailbreak), and a fourth suggests the administration simply disagreed with Anthropic’s corporate culture and political alignment.
Not yet confirmed
The precise legal basis for the export control order has not been publicly explained by the administration.
It remains unclear how Anthropic could prevent all jailbreaks of its models, as independent cybersecurity experts view guardrails as a stopgap measure.
Whether the order was triggered by a specific group’s actions (e.g., a China-linked group) or by general concerns has not been confirmed across all sources.
The exact role of other tech executives, such as Amazon CEO Andy Jassy, in raising concerns is only reported by one source.
Key figures
Anthropic, Dario Amodei (CEO of Anthropic), Donald Trump (US president), Sean Cairncross (National Cyber Director), Andy Jassy (Amazon CEO), Dean Ball (former White House AI adviser)
Sources: Wired, Ars Technica, The Verge
