AI RESEARCH

Intent Laundering: AI Safety Datasets Are Not What They Seem

arXiv CS.LG

ArXi:2602.16729v3 Announce Type: replace-cross We systematically evaluate the quality of widely used adversarial safety datasets from two perspectives: in isolation and in practice. In isolation, we examine how well these datasets reflect real-world adversarial attacks based on three defining properties: being driven by ulterior intent, well-crafted, and out-of-distribution.