In a significant development last Friday, Anthropic announced the suspension of global access to two of its AI models, Claude Fable 5 and Mythos 5, following reports of security concerns linked to discussions involving Amazon CEO Andy Jassy. According to sources, Jassy alerted Treasury Secretary Scott Bessent and other government officials that Amazon’s researchers had utilized Anthropic’s Claude Fable 5 to extract data that could potentially facilitate cyberattacks.
In light of these revelations, the government imposed an export control ban on the affected models. An Amazon spokesperson clarified that while the company frequently engages with government entities to discuss potential security risks, it does not disclose the specifics of such conversations. The spokesperson also noted that Amazon Web Services (AWS) would be impacted by the restrictions stemming from the model cutoff.
The situation escalated as further reports from The Information and Reuters indicated that Amazon, which is a significant investor in Anthropic, had conveyed its worries regarding the security implications of Anthropic’s models. This underscores the tension between innovation in AI and the need for robust cybersecurity measures.
David Sacks, who previously served as Trump’s AI czar and currently co-chairs the President’s Council of Advisors on Science and Technology, provided additional context about the discussions. He stated that a credible partner with ties to both Anthropic and the U.S. government had raised concerns regarding a “jailbreak” vulnerability associated with the models. According to Sacks, federal authorities requested Anthropic CEO Dario Amodei to address the issue by rectifying the jailbreak or withdrawing the models entirely; however, Amodei reportedly declined to comply with the request.
In response to the ongoing situation, Anthropic released a blog post emphasizing that the capabilities raising governmental apprehensions are already present in other models that are publicly accessible. This statement aims to mitigate the fears surrounding its technology while highlighting the existing vulnerabilities in the broader AI landscape.
As the dialogue around AI safety continues, both Anthropic and Amazon find themselves navigating the complex intersection of technological advancement and national security concerns. The future of these models and the nature of their deployment now hangs in the balance, with stakeholders on both sides closely monitoring the unfolding situation.
