3 papers
cs.CL2026
ContextLift: Breaking the Spurious Correlation Behind Conditional Emergent Misalignment
Florian Mai, David Kaczér
Finetuning a language model on narrowly misaligned data can make it broadly misaligned. Interleaving generic safety data is the most effective known in-training defense, but the mi…
cs.CL2026
Data Attribution of Emergent Misalignment with Persona Features
Clemens Vetter, David Kaczér, Lucie Flek +1
Emergent misalignment (EM) is the phenomenon where fine-tuning a language model on a narrow task leads to harmful behavior in unrelated domains. A leading mechanistic account attri…
cs.LG2025
In-Training Defenses against Emergent Misalignment in Language Models
David Kaczér, Magnus Jørgenvåg, Clemens Vetter +4
Fine-tuning lets practitioners repurpose aligned large language models (LLMs) for new domains, yet recent work reveals emergent misalignment (EM): Even a small, domain-specific fin…