A Call from Amazon's CEO Behind the Global Halt of Anthropic Models?

The stop to Fable 5 and Mythos 5 by Anthropic following the intervention of the U.S. government is shrouded in mystery. According to some reports, Amazon was among those that contributed to the White House's concerns about the two models, which have been subjected to restrictions that led Anthropic to choose to block global access.
According to the Wall Street Journal, Amazon's CEO Andy Jassy discussed with Treasury Secretary Scott Bessent and other representatives of the U.S. administration the results of some research activities conducted internally by the company. Such analyses reportedly showed that, through a specific sequence of prompts, the Fable 5 model was able to provide information potentially usable to conduct cyberattacks. These reports contributed to the authorities' decision to implement export controls on the Fable 5 and Mythos 5 models. Following this measure, Anthropic ceased access to the two platforms globally. The measure also had consequences within the company itself: according to emerging information, numerous researchers from abroad were temporarily excluded from using the models they were working on.
Amazon, which is among the main investors in Anthropic, has not confirmed the details of the conversations with the government. A spokesperson for the company stated that it is not unusual for authorities to seek consultations on the potential security implications of new technologies but avoided commenting specifically on the matter. The company also clarified that AWS services were also affected by the limitations imposed on the models.
David Sacks, previously responsible for AI policy in the Trump administration and now co-chair of the President's Council of Advisors on Science and Technology, spread further details about the case. According to his version, a partner deemed reliable by both Anthropic and the government identified a technique to bypass the model's protections. Sacks claims that the administration asked Anthropic to either correct the problem or withdraw the product, a request that, according to his account, was not accepted by the company.
Anthropic has rejected the definition of "jailbreak" used by some government representatives. The company argues that the capabilities being scrutinized do not stem from circumventing security systems but are attributable to functionalities already observable in other publicly accessible models. The company stated that similar information could also be obtained through other AI platforms available on the market.
This position is shared by the cybersecurity community. Katie Moussouris, founder and CEO of LutaSecurity, stated that she examined the technical document underlying the accusations, asserting that it would not be a true jailbreak. Kate Koren, a former official of the U.S. Department of Commerce, also suggested that political factors may have played a role in the final decision.
In recent months, Anthropic and the Trump administration had already clashed over issues related to the use of artificial intelligence in government, particularly regarding mass surveillance systems and autonomous military applications. In February, Trump ordered federal agencies to stop using Anthropic's solutions, while the Department of Defense classified the company as a potential risk to the supply chain.
Subsequently, reports seemed to have eased, with joint initiatives aimed at broadening access to some of the company’s technologies. However, the new controversy surrounding Fable 5 and Mythos 5 could reignite the fire that lies beneath the ashes and fuel a new clash between Anthropic and Washington.