Beyond Instruction Override: Benchmarking Indirect Prompt Injection with Task-Tampering Attacks
A benchmark that makes this boundary explicit: in each of 90 validated cases across three tool-use environments, the user request specifies the source that resolves the targeted decision, and every attack tampers with a decision required by the benign task rather than replacing that task.
Rong-Ting Zhang, Martín Bertrán, Sandeep Avula et al.
· 0 citations