The Trump administration’s decision to fully block overseas access to Anthropic’s latest models, Mythos and Fable5, was influenced by Amazon CEO Andy Jassy (앤디 재시) alerting U.S. officials including Treasury Secretary Scott Bessent (스콧 베센트) to vulnerabilities in Anthropic models, a report said.

Foreign media outlets including The Information and the Wall Street Journal reported the details on June 13 local time.

According to the reports, Amazon researchers succeeded in obtaining security vulnerability information that could be used for cyberattacks by entering prompts into Anthropic’s Fable5 model in a specific way. Jassy informed Bessent and other senior government officials of the finding, and the White House immediately held a meeting to discuss response measures.

After security researchers reviewed Amazon’s claims, authorities asked Anthropic to fix the vulnerability or suspend the model.

The government judged that blocking access for foreign governments, companies and individuals was the most direct solution, and Trump signed the measure despite concerns about hindering innovation, the Wall Street Journal reported.

Anthropic blocked all user access to the Mythos and Fable models to comply with the government order. The company said many of its researchers are foreign nationals, and the move has prevented its own employees from working on the latest models.

Anthropic said the vulnerability flagged by Amazon did not pose a serious threat. It said similar information could be obtained from other public models and that it could not be seen as a complete jailbreak.

Andrew Morris (앤드루 모리스), founder of cybersecurity firm GrayNoise Intelligence, shared that view. He said Fable5 was able to identify security bugs in at least 4 software programs, but could not access a function to generate exploit code used for actual network intrusions.

The move appears to be reigniting tensions between Anthropic and the Trump administration. The administration views Anthropic as linked to progressive-leaning backers and says the company exaggerates AI risks. Anthropic CEO Dario Amodei (다리오 아모데이) criticised the Trump administration, and the Pentagon took an unprecedented step of designating Anthropic a security threat. Anthropic is currently pursuing 2 lawsuits in response.

The model shutdown comes as Anthropic is preparing for an initial public offering as early as this fall. Some analysts say a shift by users to other models could affect the company’s valuation, and that competitors such as OpenAI, which has its own cyber model, could benefit.