Anthropic Redeploys Claude Fable 5 After US Export Controls Lifted

Anthropic's Claude Fable 5, its most capable generally available AI model, was redeployed globally on July 1, following the lifting of US export controls that had suspended its access for nearly three weeks. The emergency controls, imposed on June 12, were a direct response to an Amazon research report detailing a method to bypass Fable 5's safeguards, enabling it to identify software vulnerabilities and even generate exploit code. This marked the first instance of US export control authorities being used to restrict AI software from public access.

In response, Anthropic has integrated a new, stricter cybersecurity classifier designed to block the reported bypass technique in over 99% of cases. Requests flagged by this enhanced safeguard are now automatically routed to Claude Opus 4.8, ensuring continuity of service while mitigating potential misuse. While this new classifier may lead to more frequent flagging of benign coding and debugging tasks, its effectiveness has been endorsed as "extraordinarily strong" by researchers from the US Department of Commerce's Center for AI Standards and Innovation (CAISI). The lifting of controls also came with binding federal conditions, including proactive security monitoring and mandatory reporting of malicious use to the Commerce Department.

This episode underscores a critical juncture in AI governance, highlighting the tension between rapid innovation and responsible deployment. The fact that Anthropic's own testing revealed similar vulnerabilities in less capable models, including OpenAI's GPT-5.5 and Anthropic's own Opus 4.8, suggests the challenge of "jailbreaks" is systemic, not unique to Fable 5's advanced capabilities. Anthropic's proactive proposal for a shared industry framework to score jailbreaks, co-developed with Amazon, Microsoft, and Google, offers a promising path toward standardized AI safety protocols. However, the 18-day freeze revealed a chaotic regulatory landscape and the potential for unilateral action to disrupt global AI access, posing significant implications for international collaboration and the future of frontier AI development.