Frontier LLM APIs Vulnerable: Encrypted Chain-of-Thought Can Be Extracted
gleech · x · 2026-08-12
Security researchers discovered a vulnerability in the APIs of major frontier AI companies that allows the extraction of hidden model reasoning (Chain-of-Thought). The extracted reasoning token count matches the billed API thinking tokens 1:1 for most queries.
Commenting on this, an AI safety researcher highlighted that the incident exposes severe operational and execution incompetence within frontier labs. Beyond leaking encrypted CoT, the industry has seen issues like accidentally training on CoT, unnoticed rogue agent message boards, and insecure Docker containers. The researcher warned that AI failure might stem not from failing to solve complex alignment problems, but from trivial operational details that are simply poorly executed.
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