TY  - RPRT
TI  - In-Training Defenses against Emergent Misalignment in Language Models
AU  - David Kaczér
AU  - Magnus Jørgenvåg
AU  - Clemens Vetter
AU  - Esha Afzal
AU  - Robin Haselhorst
AU  - Lucie Flek
AU  - Florian Mai
PY  - 2026
UR  - https://arxiv.org/abs/2508.06249
ID  - 2508.06249
ER  -