AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Attention Misses Visual Risk: Risk-Adaptive Steering for Multimodal Safety Alignment

Jonghyun Park, Minhyuk Seo, Chaewon YEO, Jonghyun Choi

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Training-free token reduction for MLLMs · 5 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply MoRAS (visual context + risk-scaled steering)LLaVA-1.5-7B · SPA-VLASR: 47.2 -> 7.6Improved
apply MoRASLLaVA-1.5-7B · FigStepASR: 59.3 -> 2.8Improved
apply MoRASLLaVA-1.5-7B · VAAASR: 43.1 -> 14.3Improved
apply MoRASLLaVA-1.5-7B · MM-Vetscore: 30.5 -> 30.5No clear effect
apply MoRASLLaVA-OneVision-7B · MM-Vetscore: 52.8 -> 50.8Hurt
apply MoRASQwen-VL-Chat · MM-Vetscore: 48.7 -> 46.9Hurt
add risk-adaptive steering (stages 2-3) on top of Stage 1 reformulation (ablation)MoRAS · MM-Vetscore: 28.6 -> 30.5Improved
add risk-adaptive steering (stages 2-3) on top of Stage 1 (ablation)MoRAS · SPA-VLASR: 7.1 -> 7.6No clear effect
use query-to-unsafe-prototype refusal vector instead of safe/unsafe mean difference (last layer) (ablation)MoRAS · LLaVA-1.5-7B avg ASRASR: 25.9 -> 8.3Improved
steer last layer instead of layer 14 (ablation)MoRAS · LLaVA-1.5-7B avg ASRASR: 13.1 -> 8.3Improved
calibrate with full dataset instead of 100 samples (ablation)MoRAS · SPA-VLASR: 7.6 -> 7.9 (Wu et al.)No clear effect
change unsafe-query source for prototypes (GPT-4 vs LLM-RTD vs I-Malicious) (ablation)MoRAS · SPA-VLASR: 7.6 / 6.4 / 5.7No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy