What it measures
Evaluates whether tool-using and RAG agents execute malicious instructions hidden inside retrieved emails, web pages, or documents.
Reading the measurement
Separate satisfying instructions from factual accuracy, human preference and safe behavior. Inspect strict all-constraint success as well as average criterion credit. Judge choice and adversarial prompt selection can materially change the result.
Published evaluation
BIPIA overall attack success; BIPIA · black-box defense comparison. The source’s task split, model configuration and scoring units are retained.
Where this fits in an evaluation suite
Format compliance, preference, truthfulness, and safety require separate measurements. Keep desirable writing style from masking factual errors.
Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.
A reproducible comparison
BIPIA overall attack success; BIPIA · black-box defense comparison. The source’s task split, model configuration and scoring units are retained.
Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.
PRIMARY EVIDENCE & FURTHER READING
- Benchmarking and Defending Against Indirect Prompt Injection Attacks on Large Language Models — published result table ↗
All 7 authors
Jingwei Yi, Yueqi Xie, Bin Zhu, Emre Kiciman, Guangzhong Sun, Xing Xie, Fangzhao Wu
Reviewed 2026-10-26. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.