Researchers Extract Hidden Reasoning from Frontier Models via API Vulnerability

maksym_andr · x · 2026-08-12

Researchers have discovered a method to extract the hidden reasoning processes of frontier AI models by exploiting a vulnerability in their APIs. They verified that the extracted reasoning token count matches the billed API thinking tokens on a 1:1 basis for most queried prompts.

This reveals a significant security flaw where internal chains of thought meant to be hidden can be easily extracted. Meanwhile, related discussions highlight that Tübingen has emerged as one of the world's premier hubs for academic AI safety research, supported by multiple cross-institutional groups.

Related event: Encrypted Chain-of-Thought in Closed-Source LLMs Proven Vulnerable to Theft(16 posts)→

Original post →

More from Safety

Safety channel →