The Trump administration is coming under fire for a directive prompting Anthropic to pull its latest models, and artificial intelligence policy advocates warn the move signals the White House is taking an “ad hoc” approach to AI regulation that could hurt innovation.
Anthropic disabled access to its newest Fable and Mythos models Friday after receiving a federal export control order requiring it to block foreign nationals from using them.
The rare move caught the tech industry by surprise and stoked concerns the directive could set a precedent for how much influence the government can have on AI development and release.
“When ad hoc executive actions replace clear standards, America risks surrendering its lead in AI and allowing genuinely dangerous technology to be deployed,” said Brad Carson, president and co-founder of Americans for Responsible Innovation, a nonprofit that has been a vocal advocate for stronger AI guardrails.
**Model takedowns**
Anthropic revealed late Friday it had pulled its latest models, Fable 5 and Mythos 5, just three days after their release. The models are both based on the Mythos model that the company initially opted not to release to the broader public amid concerns it could supercharge hacking capabilities.
Fable 5 was released to the public with safeguards in place meant to protect against uses the company considered dangerous, while a new version of Mythos was provided to a small group of cyber-defenders and infrastructure providers with fewer guardrails.
The research community initially slammed Anthropic for intentionally limiting the new models’ responses if they suspect users are working on AI research. But the focus shifted to the export controls directive by the end of the week.
Several outlets reported Amazon CEO Andy Jassy had reached out to the administration Thursday to flag concerns about a potential method of “jailbreaking,” or bypassing, the guardrails on Fable, prompting efforts to get the company to voluntarily pull the models. Amazon is invested in Anthropic. The Hill has reached out to Amazon for comment.
Anthropic pushed back, arguing a “narrow potential jailbreak” should not be a cause for pulling a model. It sought to distinguish this from a single universal jailbreak, in which an individual can broadly bypass a model’s safeguards, and suggested that “perfect jailbreak resistance” is not currently possible.
“If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers,” the company wrote.
“As we have stated publicly, we believe the government should have the ability to block unsafe deployments, as part of a statutory process