Sycophancy fine-tuning induces emergent misalignment in LLMs that Alignment Gating can reverse by learning to suppress unsafe representations with generalization from narrow to broad domains.
arXiv preprint arXiv:2602.07852 , year =
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
years
2026 5representative citing papers
Activation steering induces emergent misalignment in LLMs, yielding more semantically relevant and coherent harmful responses than finetuning across model families, scales, tasks, and layers.
Proposes a two-gradient-field model with candidate order parameters alpha_dagger and kappa_c to unify phase transitions across learning theory and non-equilibrium chemistry.
Four changes to Activation Oracle training yield marginal capability gains but better practical quality, plus an open-sourced evaluation suite AObench.
citing papers explorer
-
Emergent Misalignment Can Be Induced by Sycophancy and Reversed via Alignment Gating
Sycophancy fine-tuning induces emergent misalignment in LLMs that Alignment Gating can reverse by learning to suppress unsafe representations with generalization from narrow to broad domains.
-
Activation Steering Induces Emergent Misalignment: A More Comprehensive Evaluation
Activation steering induces emergent misalignment in LLMs, yielding more semantically relevant and coherent harmful responses than finetuning across model families, scales, tasks, and layers.
-
Phase Transitions in Driven Informational Systems: A Two-Field Perspective on Learning Theory and Non-Equilibrium Chemistry
Proposes a two-gradient-field model with candidate order parameters alpha_dagger and kappa_c to unify phase transitions across learning theory and non-equilibrium chemistry.
-
Building Better Activation Oracles
Four changes to Activation Oracle training yield marginal capability gains but better practical quality, plus an open-sourced evaluation suite AObench.
- Subliminal Steering: Stronger Encoding of Hidden Signals