According to Reuters, Amazon CEO Andy Jassy (photo) has expressed concerns to the U.S. government about the security risks posed by Anthropic’s latest AI models. Those discussions took place shortly before the Trump administration intervened and restricted the use of the models worldwide.
Anthropic announced on Friday that it would disable access to its Fable 5 and Mythos 5 models worldwide. The move was prompted by an order from the U.S. government, which fears that the security measures surrounding the models could be circumvented. As a result, the systems could potentially be used to detect software vulnerabilities.
The measure is notable because it applies not only to countries considered strategic adversaries by the United States, but to all foreign users. According to Anthropic, the company was therefore required to take the models offline worldwide.
Amazon Involved in Discussions
According to a Reuters source, Amazon CEO Andy Jassy was among a group of technology leaders who met with government representatives this week to discuss the potential risks posed by advanced AI models. Amazon declined to confirm whether such discussions actually took place.
However, a spokesperson stated that governments regularly seek advice from major cloud providers on potential security issues. The company says it does not disclose information about such contacts.
Amazon’s involvement is notable because the company is one of Anthropic’s key investors and cloud partners. The AI startup runs a large portion of its infrastructure on AWS and has previously received billions of dollars in investments from Amazon.
Security Concerns Surrounding Fable 5
Anthropic released Fable 5 last week, a public version of its advanced Mythos model. The company stated that additional security measures had been implemented to prevent misuse and cyberattacks.
However, according to Anthropic, the U.S. government has pointed out a method by which that protection can be circumvented. The company acknowledges the existence of such a technique but emphasizes that it can only be used to find limited vulnerabilities and that similar possibilities also exist in other publicly available AI models.
The U.S. government apparently saw things differently. According to Anthropic, the company was ordered to stop granting foreign users access to Fable 5 and Mythos 5. Consequently, it decided to discontinue both models.
Discussion on export regulations
The restrictions were imposed through export control measures. David Sacks, President Trump’s advisor on technology and science, stated that the government took this step only after determining that Anthropic had not taken sufficient measures to resolve the security issue.
Sacks expressed the expectation that the restrictions could be lifted once Anthropic implements additional security measures.
Not everyone understands the chosen approach. Various experts who generally support export controls on advanced AI technology describe the measure as exceptionally broad. They point out that researchers and employees from friendly nations such as Canada and the United Kingdom are also affected by the restrictions.