Amazon flags Anthropic model risks to US officials
Reports say concerns preceded export controls that cut off foreign access, a major investor helps trigger restrictions on its own supplier
Images
Image Credits:Bruce Bennett / Getty Images
techcrunch.com
techcrunch.com
Dario Amodei, co-founder and chief executive officer of Anthropi.
techcrunch.com
Andy Jassy the CEO of Amazon speaks at the ceremonial ribbon cutting prior to tomorrow's opening night for the NHL's newest hockey franchise the Seattle Kraken at the Climate Pledge Arena on October 22, 2021 in Seattle, Washington
techcrunch.com
techcrunch.com
A person hangs up posters on a wall. We only see the person's hands and part of their right arm. The poster centered in the photo says "Build w/ Claude" referring to Anthropic's AI technology.
motherjones.com
Anthropic disabled access to its Fable 5 and Mythos 5 models for customers worldwide after the US government ordered restrictions on foreign access, according to TechCrunch and Mother Jones. The move followed reporting that Amazon chief executive Andy Jassy had warned senior US officials that researchers at Amazon were able to use one of Anthropic’s models to obtain information potentially usable in cyberattacks. The Wall Street Journal said Jassy raised the issue with Treasury Secretary Scott Bessent and other officials.
The sequence matters because Anthropic is not only a high-profile AI supplier; it is also a company backed by the same firms that buy and integrate its models. When a leading investor and customer can brief regulators about a competitor-supplier’s weaknesses, the boundary between market competition and national-security policy becomes hard to trace. TechCrunch reported that Reuters and The Information also described Amazon communicating concerns about Anthropic’s security, while an Amazon spokesperson told the Journal it is common for governments to seek Amazon’s counsel on risks, without confirming what was discussed.
The administration’s public rationale has been national security, but the mechanism is commercial. Anthropic complied by switching off access to the two models for all customers, Mother Jones reported, after being ordered to suspend foreign nationals, including employees, from using its most advanced products. That is an export control enforced through a private company’s product dashboard, not a customs checkpoint. For customers outside the US, the practical message is that “sovereign AI” debates can end with a login screen.
The accounts also describe a dispute inside the US itself about how far these controls should go. David Sacks, described by TechCrunch as the former AI czar under President Trump and current co-chair of the President’s Council of Advisors on Science and Technology, said a “highly credible trusted partner” of both Anthropic and the US government came forward with information about a jailbreak. Sacks claimed the administration asked Anthropic chief executive Dario Amodei to fix the issue or de-deploy the model, and that Amodei refused. Mother Jones, meanwhile, notes Anthropic has marketed itself as an “ethical AI” company and has said some government demands—such as mass surveillance of Americans and fully autonomous weapons—would violate its safeguard policies.
The episode lands as Anthropic prepares for an initial public offering, according to Mother Jones, and after a period in which the company has tried to position its guardrails as a competitive advantage. But those guardrails are now being evaluated in a different forum: by officials who can impose restrictions without detailing the underlying evidence. For developers, the risk is not only that models can be misused, but that access can be revoked overnight, with the decision shaped by actors who may also profit from the outcome.
Anthropic’s most advanced models were available to customers until the company disabled them to comply with the directive. The restriction did not arrive as a new product launch or a patch note, but as a service interruption tied to a government order that, so far, has not been publicly explained in detail.