Amazon security research led to White House Anthropic ban
▼ Summary
– A Wall Street Journal report states that Amazon’s cybersecurity research and CEO Andy Jassy’s White House discussions triggered export controls that led Anthropic to block access to its Fable 5 and Mythos 5 models.
– Amazon’s research claimed it could prompt Fable 5 to provide information useful for cyberattacks, and after Jassy shared findings, the government restricted use by foreign nationals.
– Anthropic disputed the “jailbreak” characterization, arguing similar vulnerabilities exist in other public models like GPT 5.5, a view backed by some security researchers.
– The Trump administration and Anthropic have a history of conflict, including over the company’s refusal to allow AI use for mass surveillance or autonomous weapons, leading to federal bans and a supply chain risk designation.
– Despite recent efforts to reconcile and expand access to Mythos, the government and Anthropic appear headed for renewed conflict.
A new report from the Wall Street Journal reveals that an export control directive from the White House, which forced Anthropic to restrict access to its Fable 5 and Mythos 5 models, was fueled by cybersecurity research conducted by Amazon. The directive ultimately prevented foreign nationals, including many of Anthropic’s own researchers, from using the company’s products.
According to the report, Amazon’s research paper demonstrated that, through a specific sequence of prompts, it could get Fable 5 to generate information useful for launching cyberattacks. Amazon CEO Andy Jassy reportedly shared these findings with the White House, which then moved to block the models’ use by non-U. S. citizens. Amazon has not yet responded to requests for comment on the matter.
This decision created a deeply ironic problem: a significant portion of Anthropic’s research team is foreign-born, meaning they were suddenly locked out of the very AI systems they helped build. Anthropic has pushed back against the government’s classification of the issue as a “jailbreak,” arguing that similar vulnerabilities are present in other publicly available models, including GPT 5.5. Some security experts agree. Katie Moussouris, founder of LutaSecurity, posted on BlueSky that she has reviewed the paper and declared, “It’s not a jailbreak.” Former Commerce Department official Kate Koren suggested to the WSJ that the White House’s existing friction with Anthropic may have played a role in the aggressive response.
Tensions between Anthropic and the Trump administration are nothing new. The company has long refused to let its AI be used for mass surveillance of Americans or to power lethal autonomous weapons. In February, President Trump ordered federal agencies to stop using Anthropic’s AI, and within hours, Secretary of Defense Pete Hegseth designated the company a supply chain risk. While the two sides appeared to have recently mended fences and collaborated on expanding access to Mythos, this latest clash suggests a return to open conflict.
(Source: The Verge)
