The authors used an agent framework to retrieve public material, extract processing methods, remove duplicates and generate executable code.
Web-derived image edits expose uneven weaknesses in AI image detectors
Li and colleagues turn public web material into 156 executable tools and find detector-specific evasion weaknesses across six AI-image detectors.
Research Lab
Mingjie Li · Yukun Jiang · Yicong Tan · Yang Zhang
CISPA Helmholtz Center for Information Security
Research Digest··2 min read
Li and colleagues introduce WildFakeToolBench, a collection of CPU-compatible image-processing tools derived from public forums, repositories and tutorials.
§