Alignment Forum4 min read
Automating subtle emergent misalignment research
Authors: Oscar Balcells Obeso, Richard Qi, Monte MacDiarmid, Evan HubingerAbstractRed-teaming by humans is a vital part of model safety testing, but it may struggle to keep up as models improve.
By evhubAlignment Forum

Read the full story
Originally published by Alignment Forum on Nov 3, 2026.
Comments
Sign in to join the discussion.