Colleagues, I’d like to highlight the return of Claude Fable 5 after the lifting of US export restrictions

Colleagues, I’d like to highlight a development in cybersecurity and AI.

Anthropic is once again rolling out Claude Fable 5 globally after the U.S. Department of Commerce lifted export restrictions.

The pause was driven by a jailbreak scenario: the model was able to bypass part of its safeguards, creating a dual-use risk.

The company says it has introduced a new filter and now blocks this technique in more than 99% of cases. For contentious prompts, some tasks are routed to a weaker model.

I believe this is not only about the service’s return, but also about how we assess the risks of frontier AI models.

How do you assess the balance between safety and accessibility in such systems?

#cybersecurity #AI #Anthropic #jailbreak