Poster · Paris

Neural Chameleons: Language Models Can Learn to Hide Their Thoughts from Unseen Activation Monitors

Name: Neural Chameleons: Language Models Can Learn to Hide Their Thoughts from Unseen Activation Monitors
Start: 2026-12-11T11:30:00+01:00
End: 2026-12-11T13:30:00+01:00
Location: Paris Poster Hall, Paris

When: Fri 11 Dec 2026, 11:30–13:30 CET
Where: Paris Poster Hall
Session: Paris Poster Session 5
Oral: Oral of this paper
Authors: Max McGuinness, Alex Serrano Terre, Luke Bailey, Scott Emmons
Organisations: Anthropic, MATS, Stanford University, Google DeepMind
QRS classification: Language models

Open in schedule Official page