safety security
Green Words Made the Vision Model Read the Sentence Differently
Subtle color and contrast changes shifted sentiment and visual-question answers even when the rendered words stayed the same.
Summary
Subtle color and contrast changes shifted sentiment and visual-question answers even when the rendered words stayed the same.
Stealth Visual Prompts alter the styling of text rendered as an image without changing its words. Across the reported experiments, coloring positive words green moved sentiment predictions in a positive direction and sometimes obscured negative content; reducing contrast increased reliance on salient visual cues and produced more wrong answers. The study shows a presentation-layer vulnerability in tested vision-language models, not a claim that every model maps green to approval.
Why it matters
Subtle color and contrast changes shifted sentiment and visual-question answers even when the rendered words stayed the same.
Limits and context
- The study shows a presentation-layer vulnerability in tested vision-language models, not a claim that every model maps green to approval.
Key claims
Subtle color and contrast changes shifted sentiment and visual-question answers even when the rendered words stayed the same.
Qualification: The study shows a presentation-layer vulnerability in tested vision-language models, not a claim that every model maps green to approval.
Evidence: source-2026-08-17-015
Sources
- arXiv preprint 2608.14286arXiv · primary research
Corrections
No corrections have been recorded for this story.