After back-to-back accusations of Chinese model distillation, Anthropic releases Fable 5 Reflection Summary

According to Perceiving Motion monitoring, Fable 5 no longer displays the Thinking Summary on its web, desktop, and mobile interfaces. The model is still running inference in the background, but users can only see the final answer. Anthropic has not yet explained the reason, but this is most likely the latest measure in anti-distillation.

In February of this year, Anthropic accused DeepSeek, DarkSide of the Moon, and MiniMax of using around 24,000 fake accounts to interact with Claude over 16 million times. DarkSide of the Moon was specifically accused of extracting and reconstructing Claude's reasoning trajectory.

In June, Anthropic again accused Ali and affiliates of a Thousand Questions of using nearly 25,000 fake accounts to interact with Claude over 28.8 million times. In July, Anthropic's National Security Director Tarun Chhabra publicly implicated SmartSpectrum for the first time, stating that GLM-5.2 had distilled Claude and the OpenAI model. After Kimi K3 was released, White House Chief Technology Officer Michael Kratsios accused DarkSide of the Moon of using Fable to develop K3.

Fable 5 has integrated a "Reasoning Extraction" classifier, specifically blocking requests to extract Thinking Summaries. Now, Anthropic has directly hidden the Thinking Summary in the chat interface, making it harder for outsiders to access anything other than the final answer. This makes it more difficult to collect in bulk the model's problem-solving process for distillation. The downside is that even regular users cannot see what the model is working on, making it harder to detect deviations in long tasks, correct them promptly, or halt them.