cd/entity/ViSafe-Eval· home› entities› ViSafe-Eval
grep -l @visafe-eval /news/*.json | wc -l → 1

ViSafe-Eval

mentions 1 type Organization feed RSS

// recent coverage 1 mentions

17:46
2026-09-01
promptcube3.com
ai-safety

Visual inputs are bypassing LLM safety filters in ways that

Researchers analyzing 10 vision-language models (VLMs) found that visual inputs bypass safety filters because text-based refusal relies on a tiny cluster of about 88 neurons (less than 0.01% of total)…

// co-occurs with top 2 entities
ViSafe-Detect 1 jiaxuan-li.github.io 1
// topics top 4 topics
ai safety 1 ai research 1 large language models 1 computer vision 1