TY  - RPRT
TI  - Why Summaries Turn Neutral: Policy Attribution for Sentiment Drift in Reinforcement Learning from Human Feedback
AU  - Mikhail Krasitskii
AU  - Alexander Gelbukh
AU  - Olga Kolesnikova
AU  - Grigori Sidorov
PY  - 2026
UR  - https://arxiv.org/abs/2608.15530
ID  - 2608.15530
ER  -