The authors built BrowseSafe++, extending BrowseSafe from 14,719 to 38,675 attack/control page pairs.
BrowseSafe++ benchmark evaluates prompt injection detection across real web contexts
A dataset of 38,675 page pairs with real-world injections and four views enables robust evaluation and training of compact detectors.
Academic
Yuetian Chen · Kaiyuan Zhang · Yuntao Du · Thomas Li · Mark Tenenholtz · Kyle Polley · +3 more
Perplexity · Purdue University · Rutgers University
Research Digest··2 min read
Yuetian Chen and colleagues introduce BrowseSafe++, a benchmark for indirect prompt injection detection that includes 32,187 pairs from production websites, each with HTML, Markdown, accessibility tree, and screenshot views.
§