Researchers Show Encrypted Chain-of-Thought Can Be Stolen via Two-Step API Calls

The paper "Stolen Thoughts" reveals that APIs from OpenAI, Anthropic and Google return encrypted chain-of-thought blocks that can be decrypted with two API calls. Anthropic has responded with a preserved thinking feature to block such attacks.

2026-10-13 ~ 2026-10-14 · 2 related posts

• Stolen Thoughts: Two API Calls Extract Frontier Model Reasoning; Anthropic Rolls Out Preserved Thinking — zainhas · 2026-10-13
• Encrypted Chain-of-Thought Can Be Decoded: Reasoning Theft Attack on Top LLM APIs — matthew_d_green · 2026-10-14