Counterfactual Neighborhood Risk Fields: Evidence-Grounded Selective Safety for Vision-Language Models

arXiv:2610.20679v1 Announce Type: new Abstract: Representation-based safety defenses for vision-language models (VLMs) often learn to separate harmful from benign inputs, but separation alone does not establish what a risk score means or when it should be trusted. We introduce Counterfactual Neighbo

arXiv cs.AIFreitag 06:00 · 16.10.1 Min. Lesezeit