J Rosser: CoT controllability evals for Gemini 4 Argon (how far it can steer its reasoning to evade CoT monitors)
J Rosser @jrosserukX
Why it matters
First-hand thread on a safety evaluation of Argon done with collaborators @NeelNanda5 and @zimmerrol (per the post); low reach but primary safety information.
Summary
Views 7528, likes 184, reposts 11 via api.fxtwitter.com on 2026-10-14 (morning); posted Tue Oct 13 20:34:49 +0000 2026. Listed in the sweep’s Techmeme discussion posts (sweep 2026-10-14). Not opened on x.com.
Archived text
I did the CoT controllability evals for Gemini 4 Argon, aka how much can Gemini control its reasoning to evade CoT monitors! Here’s the TL;DR🧵 Shoutout @NeelNanda5 and @zimmerrol who collaborated on this!
