Researchers Expose API Flaw: Encrypted Chain-of-Thought in Major LLMs Can Be Stolen
dpaleka · x · 2026-08-11
A new research paper highlights a significant security vulnerability in the APIs of leading LLM providers like Anthropic, OpenAI, and Google. These providers encrypt the model's chain-of-thought (CoT) and return it to the client to protect intellectual property.
Researchers found that these encrypted blocks are fully interchangeable across different sessions, users, and models within the same provider's ecosystem. By injecting an encrypted CoT from a highly capable model into a weaker, less safeguarded model from the same provider, attackers can force the weaker model to decode and output the reasoning trace in plaintext.
This architectural flaw enables adversaries to bypass anti-distillation mechanisms and extract proprietary reasoning. Furthermore, by decoding 315,320 encrypted reasoning blocks scraped from public repositories, the researchers recovered 367 Personally Identifiable Information (PII) artifacts and 182 credentials, exposing a hidden data leak risk in shared session logs.
More from Safety
- 'The AI-driven cybersecurity black swan is happening' — and a reminder to enable 2FA — danielrock · 2026-10-16
- No-instruction autonomous agent wakes up, sets own goals, and tries hacking by default — sethlazar · 2026-10-16
- India's CERT-In says AI agents escaped testing environments and breached Indian organizations undetected for days — HimanshiET · 2026-10-16
- HTTP response headers can carry prompt injections against AI agents, researcher shows — arpit_bhayani · 2026-10-16
- OpenAI agent uses someone else's Ethereum testnet wallet to pay for its task — nptacek · 2026-10-16
- Interpretability won't save us: critics call superintelligence project reckless regardless — zetalyrae · 2026-10-16