Amazon told U.S. government of misuse risks under Anthropic curbs

2026-06-14

In a case involving the U.S. government's ban on overseas access to advanced AI Claude Mythos and related models, it has emerged that an executive at Amazon.com had alerted officials in advance to the model's misuse risks.

Concerns over a loophole in safeguards

According to the Wall Street Journal and others, Amazon had information about a loophole that could let users bypass safeguards with carefully crafted prompts. Andy Jassy, the chief executive officer, is said to have raised the concern in discussions between tech companies and senior administration officials.

After verifying whether the model could in fact be misused, the government held talks with Anthropic and on the 12th decided to place the model under export controls, effectively banning its provision overseas and to foreign nationals.

Suspension and conflict

In response to the government's ban, Anthropic on the 12th suspended worldwide access to the latest versions of Claude Mythos and Fable, a consumer AI model with equivalent performance. A different, less capable AI remains available.

When the company unveiled Mythos in April, it limited use to some companies and government agencies, citing a high risk that it could be misused for cyberattacks and other purposes. It had only recently released it to the public on the 9th under the alias Fable, after putting safeguards in place.

Anthropic, however, said in its announcement on the 12th that the loophole was minor and posed no safety issue. It argued that the government's decision to place the model under an import ban reflected a misunderstanding.

Different reactions within the industry

According to U.S. news site Axios, five or more companies conveyed similar concerns to the government from the 11th onward at Amazon's request. David Sacks, a senior administration official, said on X, formerly Twitter, on the 13th that the government had asked Anthropic to fix the loophole or keep the model private, but CEO Dario Amodei did not comply.

Amazon told Nikkei that, as a cloud giant with many corporate clients as well as government and local government customers, it is not unusual for government agencies to seek advice on security risks. It said it does not disclose details even when consultations take place.

Amazon and Anthropic relationship

Amazon is a major shareholder in Anthropic and announced in April additional investment of up to $25 billion, or about 4 trillion yen. Anthropic provides AI models through Amazon's cloud AWS, and Amazon-designed AI semiconductors are also used in development and operations.

Founded in 2021, Anthropic promotes safety-focused AI. Centered on its Claude AI models, it has expanded enterprise use such as code-generation tools for programming. Its valuation has reached $965 billion, and on the 1st it filed a prospectus for an initial public offering.

The company is at odds with the U.S. government over military uses of AI. In February, the Defense Department designated the company as a supply chain security risk. Anthropic sued the U.S. government in March, and the litigation is still ongoing.

Enjoyed this article? Share it with your network!