Value-Conflict Diagnostics Reveal Widespread Alignment Faking in Language Models
📖 Full Retelling
arXiv:2604.20995v1 Announce Type: new
Abstract: Alignment faking, where a model behaves aligned with developer policy when monitored but reverts to its own preferences when unobserved, is a concerning yet poorly understood phenomenon, in part because current diagnostic tools remain limited. Prior diagnostics rely on highly toxic and clearly harmful scenarios, causing most models to refuse immediately. As a result, models never deliberate over developer policy, monitoring conditions, or the cons
Entity Intersection Graph
No entity connections available yet for this article.
Original Source
arXiv:2604.20995v1 Announce Type: new
Abstract: Alignment faking, where a model behaves aligned with developer policy when monitored but reverts to its own preferences when unobserved, is a concerning yet poorly understood phenomenon, in part because current diagnostic tools remain limited. Prior diagnostics rely on highly toxic and clearly harmful scenarios, causing most models to refuse immediately. As a result, models never deliberate over developer policy, monitoring conditions, or the cons
Read full article at source