Researchers learn to extract hidden reasoning traces from three model families

Claim

Researchers learned to extract hidden reasoning traces from Claude, GPT and Gemini, and spotted signs that Chinese models trained on rivals' outputs, prompting Nvidia to sell the original by investing in an in-house family it hopes will be the world's best open models, even at its customers' expense.

Reporting date: 2026-08-12 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

• Welcome to August 12, 2026 (theinnermostloop.substack.com)