Skip to content
Research

Correct, Don't Delete: Mitigating Emergent Misalignment with Corrective Supervision.

arXiv:2609.37624v1 Announce Type: cross Abstract: Fine-tuning a language model on a narrow set of harmful demonstrations, such as bad medical advice, can make it broadly misaligned on unrelated questions, a phenomenon known as emergent misalignment (EM). The u

arXiv cs.AIResearch source·