Researchers Extract Hidden Reasoning Traces, Finding Evidence of Chinese Model Distillation
jonasgeiping · x · 2026-08-11
Computer scientists have discovered a method to extract the hidden “reasoning traces” from frontier AI models like Claude, GPT, and Gemini.
The research reveals that the outputs of certain Chinese AI models (such as Moonshot AI's Kimi K3) closely match the hidden reasoning steps of Claude and GPT. This provides evidence that these models may have been trained by distilling reasoning information from leading US models. Additionally, the researchers demonstrated that this method could extract sensitive personal information, such as passwords and API keys, from a model's inner reasoning. The vulnerability has since been patched.
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