A recent report indicates that Andy Jassy, the chief executive of Amazon.com Inc., was instrumental in the U.S. government's decision to impose an export control ban on Anthropic's sophisticated "Mythos-class" AI models. These models, designed for intricate reasoning and prolonged autonomous operations, faced scrutiny due to potential security vulnerabilities. According to sources cited by The Wall Street Journal, Jassy communicated to Treasury Secretary Scott Bessent that Amazon's research teams discovered the Fable 5 model could be leveraged for cyberattacks. This revelation prompted Anthropic to globally discontinue access to these models.
Following the government's directive, Anthropic moved on Friday to disable access to both Fable 5 and Mythos 5, citing national security imperatives that necessitated barring foreign access to these systems. Similar concerns regarding Amazon's involvement were echoed by The Information and Reuters, emphasizing Amazon's significant investment in Anthropic. In April, Anthropic pledged to invest over $100 billion into Amazon Web Services infrastructure, including Trainium chip capacity, while Amazon concurrently injected an additional $5 billion into the startup, supplementing its initial $8 billion stake. Separately, David Sacks, a former White House AI and Crypto Czar and co-chair of the President's Council of Advisors on Science and Technology, stated on X that a trusted partner of both Anthropic and the U.S. government identified a "jailbreak" in Fable during testing. Sacks claimed that despite requests from the administration to either rectify the vulnerability or decommission the model, Anthropic's CEO, Dario Amodei, declined. Anthropic had previously noted in a blog post that the capabilities drawing government concern are also present in other widely available AI models.
This incident underscores the critical balance between technological advancement and national security. It highlights the profound responsibility that technology leaders and governments share in managing the dual-use nature of advanced AI, ensuring that innovations serve humanity's progress while safeguarding against potential misuse. The swift action taken in response to identified vulnerabilities reflects a commitment to prioritizing security, fostering trust, and preventing the exploitation of powerful AI tools for malicious purposes.