ainewsblitz.com

Breaking

Researchers extract hidden reasoning from Claude, GPT, and Gemini APIs

Researchers extract hidden reasoning from Claude, GPT, and Gemini APIs
  • Research & Papers
  • Security
  • Foundation Models

Researchers have shown that encrypted chain-of-thought from leading U.S. AI APIs can be recovered at scale by replaying it through weaker models from the same provider, and they report that Moonshot AI’s open-weight Kimi K3 sometimes behaves as if it has already seen those hidden traces. The work does not prove that Chinese labs distilled American systems, but it lands in a running U.S.–China dispute over unauthorized capability extraction.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.

$20
Read this article
$29/month
Unlimited — all 27,595 articles, the full archive, and comprehension quizzes
Save 72%
$98/year
≈ $8.17/month
Unlimited, billed once a year