It has emerged that a tip from Amazon, a key investor in Anthropic, was behind the U.S. government's blanket block on foreign access to Anthropic's latest AI model.
On the 13th (local time), the Wall Street Journal (WSJ), citing multiple sources, reported that Amazon researchers notified a government official that they had found vulnerabilities in Anthropic's "Fable 5" model that could be exploited for cyberattacks, prompting the U.S. government to impose a block. A senior White House official said President Donald Trump gave the final approval.
Fable 5 was designed to block commands related to cybersecurity or biochemical weapons, but Amazon researchers succeeded in bypassing these safeguards by entering a series of prompts (commands). After U.S. government security personnel verified this, they concluded that blocking access to the model by foreign governments, corporations, and individuals was the most direct solution.
On the 12th, the U.S. government announced export control guidelines banning all foreign nationals from accessing "Fable 5" and "Mythos 5." To comply with the guidelines, Anthropic temporarily suspended all users' access to the models, and said the claims of possible bypassing were a misunderstanding while pledging to work to restore the service.
Regarding this, Amazon said, "As a cloud provider, it is not uncommon for the government to seek our advice on potential security risks," but added, "We do not share the specifics of those discussions."
The episode has brought the complex interests of Amazon and Anthropic to the surface. Since 2023, Amazon has invested a total of $13 billion (about 20 trillion won) in Anthropic as a key shareholder, while this year diversifying its AI investment portfolio by committing up to $50 billion (about 75 trillion won) to rival OpenAI.
Questions are also being raised about whether the move was a purely security judgment. Kate Corn, deputy director of the economics program at the Center for Strategic and International Studies (CSIS), said, "I acknowledge there are security concerns, but the White House's antipathy toward Anthropic likely influenced this decision."
The WSJ also noted that litigation between Anthropic and U.S. administrative bodies such as the Ministry of National Defense may have influenced the move. A senior White House official, however, denied any political background, saying, "This action is about model safety."