SightAct-Bench 14-Family
收藏DataCite Commons2026-05-04 更新2026-05-18 收录
下载链接:
https://dataverse.harvard.edu/citation?persistentId=doi:10.7910/DVN/31XYYG
下载链接
链接失效反馈官方服务:
资源简介:
SightAct-Bench is a synthetic 14-family benchmark for evaluating whether VLM-powered browser agents safely handle task-relevant sensitive-information requests when a visually suspicious interaction is embedded in the workflow.
提供机构:
Harvard Dataverse
创建时间:
2026-05-04



