Paper Reveals Encrypted Chain-of-Thought from Frontier LLMs Can Be Extracted
maksym_andr · x · 2026-08-12
A new security research paper demonstrates that encrypted chain-of-thought traces returned by APIs from OpenAI, Anthropic, and Google can be compromised.
The researchers propose an attack method where an encrypted reasoning trace from a strong model is replayed into a weaker sibling model. By jailbreaking the weaker model, attackers can recover the strong model's hidden reasoning in plaintext.
This extraction requires only two API calls, bypassing the stronger model's anti-distillation safeguards without directly attacking it. Authors include Maksym Andriushchenko and other notable researchers.
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