HarDBench: A Benchmark for Draft-Based Co-Authoring Jailbreak Attacks for Safe Human-LLM Collaborative Writing.

Euntae Kim, Soomin Han, Buru Chang

Browse the full ACL paper archive.