Reaction
Frontier AI company APIs allegedly allow extraction of hidden model reasoning
A poster says their reasoning-token counts matched billed API thinking tokens 1:1 for most prompts they queried.
TLDR
An August 11 post claims a vulnerability in the APIs of every frontier AI company can expose hidden model reasoning; its authors say their token counts matched billed thinking tokens 1:1 for most prompts they tested. An October 15 quote post claims its author's distillation paper forced labs to rewrite their stance and alleges distillation was 5–7% of Anthropic's $16.8 billion quarter before labs were prompted to crack down on it.
Combined views
—
1 Source, first seen ago
