Anthropic’s Claude Fable 5 curbs to create new hurdles for China’s AI labs

Anthropic, a US-based AI company, has imposed stricter access controls on its Claude Fable 5 model, a public version of its most advanced system, Mythos. The restrictions aim to curb misuse by Chinese AI labs, particularly in areas like cybersecurity, biology, chemistry, and frontier large language model development. When queries related to these fields are detected, Fable 5 downgrades responses to its second-best model, Claude Opus 4.8, making it harder to exploit vulnerabilities or accelerate model development. The decision comes after concerns that users could bypass safeguards in earlier versions, such as Mythos, which demonstrated an alarming ability to identify and exploit cybersecurity flaws. Anthropic’s move has drawn criticism from the global AI research community, leading the company to partially roll back enforcement. Experts, including Kyle Chan from the Brookings Institution, warn that Chinese AI developers may now face significant obstacles in leveraging Anthropic’s technology for their own advancements. Claude Fable 5 was designed to deliver high-performance capabilities in complex tasks like software engineering and scientific research. However, its built-in classifiers now actively block sensitive queries, limiting its utility in certain research areas. The restrictions have reignited discussions about balancing AI safety with open innovation, particularly as China’s AI sector continues to expand. Anthropic’s adjustments to enforcement reflect broader tensions between regulatory caution and the need for accessible AI tools. The company’s approach highlights the challenges of preventing misuse while fostering collaboration in the AI research community. Meanwhile, Chinese AI labs may explore alternative models or develop their own solutions to overcome these new barriers.