Encrypted Chain-of-Thought in Frontier LLMs Can Be Ported Across Users, Researchers Confirm
AccBalanced · x · 2026-08-12
Security researchers have confirmed significant vulnerabilities in the encrypted reasoning processes (chain-of-thought) of frontier LLMs like GPT and Claude.
- Background: In May, Matthew Green found that encrypted reasoning could be replayed outside its original context, but AI labs dismissed it as a non-threat.
- New Findings: A recent report confirms that encrypted thoughts are fully portable across different sessions, users, and even models within the same provider.
- Implications: This means a mere trickle of API usage could allow bad actors to exfiltrate model capabilities easily, bypassing safety theatrics. Additionally, the presence of Chinese/Russian/Japanese reasoning traces in these models hints at their training data origins.
More from Models
- AI Assistant Flags Normal Tax Data Update as Fraud in Finance Test — daniel-editide · 2026-08-13
- Grok Iteration Test: Version 4.6 Crushes Previous Build in Minecraft Spatial Judgment — adonis_singh · 2026-08-13
- Grok 4.6 in Action: 10 Wild Coding and 3D Generation Demos Within 48 Hours — minchoi · 2026-08-13
- Grok 4.6 Beats DeepSeek-V4-Pro Across 30 Agentic Tasks at Lower Cost — ns123abc · 2026-08-13
- Claude 3.7 Flash Now Available for Testing on Vertex AI — Big-Reason-2976 · 2026-08-13
- SenseNova-Vision: A 7B Open Model Unifying Segmentation, Depth, and 3D Reconstruction — SandyL925 · 2026-08-13