SOURCE-LINKED INTELLIGENCE

Taming CoT Obfuscation in VLMs: From Mechanistic Evidence to Activation Enforcement

arXiv · AI, language, vision and robotics · article · Sep 21, 2026 · UTC

Reinforcement learning (RL) improves reasoning in vision-language models (VLMs) but can induce chain-of-thought (CoT) obfuscation: an operational, non-intentional outcome where task reward or accuracy rises while traces become less grounded and monitorable. Prior work largely documents this decay behaviorally, leaving its representation-level correlates and actionable controls unclear. We find that template- and ground-associated activations become less separable during RL; matched interventions support the contribution of selected features to monitorability degradation. Guided by this evidenc

Read original source ↗ Open in workspace

recordType

paper

region

Global