Amazon CEO’s Warnings Precede Anthropic’s Sudden Shutdown of Two AI Models Over Security Risks
Anthropic, a leading artificial intelligence company backed by Amazon, has abruptly suspended access to its two most advanced AI models—Fable 5 and Mythos 5—after the U.S. government ordered it to block foreign users, citing national security concerns and following internal warnings from Amazon’s CEO about potential misuse [172723][172233][172418].
The shutdown, announced Friday, came after the Trump administration issued an export control directive requiring Anthropic to restrict access to the models for foreign nationals both inside and outside the United States [172418][172180][172376]. The company said it was not given specific details of the national security threat but complied by disabling the models for all users [172233][172418]. Anthropic had launched Fable 5 just days earlier for subscribers of its Claude service, promoting it as its most advanced model yet [172418].
The move follows reports that Amazon CEO Andy Jassy raised security concerns about the models before the global shutdown [172723]. Anthropic, which is backed by Amazon, suspended the models after what it described as a security review, and sources say Jassy’s worries may have accelerated the decision [172723]. Neither company has disclosed the specific risks that triggered the review [172723].
The U.S. government believes the safeguards on Fable 5 and Mythos 5 can be bypassed, potentially allowing the systems to be used to identify software vulnerabilities [172233]. Anthropic had previously shared Mythos with the U.S. government and select companies through its Project Glasswing cybersecurity initiative to test for vulnerabilities before the public release of Fable 5 [172418]. Despite those safeguards, the government’s directive led to the full shutdown [172418].
Anthropic expressed frustration with the process, stating: “We believe the government should have the ability to block unsafe developments as part of a transparent, fair, and fact-based process. This action does not adhere to those principles” [172418]. The company had previously argued that the U.S. government should have the power to shut down dangerous AI models, with CEO Dario Amodei proposing mandatory safety testing for frontier AI systems [170116].
The suspension is temporary, pending a review of security protocols [172020]. The export-control measure is the first of its kind to target specific AI models, highlighting the U.S. government’s increasingly direct approach to regulating frontier AI technology [172060].