After Anthropic decided to restrict access worldwide for two advanced artificial intelligence models, new details emerged about the reasons behind the process. According to The Wall Street Journal, Amazon CEO Andy Jassy stated that in his meetings with US Treasury Secretary Scott Bessent and other government officials, company researchers used the Anthropic Claude Fable 5 model to obtain information that can be evaluated in cyber attacks. Following these negotiations, the US administration imposed an export control ban on the Fable 5 and Mythos 5 models.
Making a statement on the subject, an Amazon spokesperson stated that it is normal for governments to seek opinions from companies about potential security risks from time to time. However, the company emphasized that it does not share the content of such conversations with official institutions. The Amazon front also drew attention to the update showing that AWS services were also affected by Anthropic’s access restriction decision.
In separate news published by The Information and Reuters, it was claimed that Amazon contacted the authorities about the security risks of Anthropic models. These claims are noteworthy because Amazon is among Anthropic’s largest investors and the two companies maintain close cooperation in the fields of cloud computing and artificial intelligence.
It is claimed that there is a difference of opinion between the US administration and Anthropic.
David Sacks, one of the co-chairs of the US President’s Council of Science and Technology Advisers, also shared his own assessment on the subject. Sacks, who previously came to prominence with his consultancy duties in the field of artificial intelligence in the Trump administration, claimed that a business partner considered reliable by both Anthropic and the US government offered a “jailbreak” method that revealed vulnerabilities in the model.
According to Sacks’ statement, the US administration requested Anthropic CEO Dario Amodei to fix the vulnerability or decommission the model. However, Sacks claimed that Amodei did not accept these demands. These claims have not been verified by Anthropic so far.
On the other hand, in the statement published on its official blog, Anthropic argued that the capabilities that the government is concerned about are not specific to its own models. According to the company, functions with similar capabilities are already available in other publicly accessible artificial intelligence models. For this reason, the company points out that the decision taken may not completely eliminate security risks.
The ability of productive artificial intelligence systems to perform more complex tasks has recently brought about security and control discussions. In particular, usage scenarios related to cyber security, biotechnology and critical infrastructures are closely monitored by both technology companies and regulatory institutions. The developments around Anthropic’s Fable 5 and Mythos 5 models also show that the debates between the security balance and access policies in the field of artificial intelligence may intensify in the coming period. Although the information available to the public is limited for now, the parties’ statements reveal that the issue has not only technical but also regulatory and political dimensions.