Learning Efficient Guardrails for Compliance

Xiaofei Wen, Wenjie Jacky Mo, Yanan Xie, Peng Qi, Muhao Chen
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:134614-134630, 2026.

Abstract

Autonomous web agents are increasingly deployed for long-horizon tasks, yet their ability to adhere to real-world policies remains critically underexplored compared to standard safety objectives. To address this gap, we introduce PolicyGuardBench, a benchmark of 60k policy-trajectory pairs designed to evaluate compliance through both full-trajectory and novel prefix-based violation detection tasks. Using this dataset, we train PolicyGuard, a lightweight guardrail model that achieves strong detection accuracy while maintaining high inference efficiency. Notably, our model demonstrates robust generalization capabilities, preserving high performance even on unseen domains. These contributions establish a comprehensive framework for studying policy compliance, showing that accurate and generalizable guardrails are feasible at small scales.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-wen26g, title = {Learning Efficient Guardrails for Compliance}, author = {Wen, Xiaofei and Mo, Wenjie Jacky and Xie, Yanan and Qi, Peng and Chen, Muhao}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {134614--134630}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/wen26g/wen26g.pdf}, url = {https://proceedings.mlr.press/v306/wen26g.html}, abstract = {Autonomous web agents are increasingly deployed for long-horizon tasks, yet their ability to adhere to real-world policies remains critically underexplored compared to standard safety objectives. To address this gap, we introduce PolicyGuardBench, a benchmark of 60k policy-trajectory pairs designed to evaluate compliance through both full-trajectory and novel prefix-based violation detection tasks. Using this dataset, we train PolicyGuard, a lightweight guardrail model that achieves strong detection accuracy while maintaining high inference efficiency. Notably, our model demonstrates robust generalization capabilities, preserving high performance even on unseen domains. These contributions establish a comprehensive framework for studying policy compliance, showing that accurate and generalizable guardrails are feasible at small scales.} }
Endnote
%0 Conference Paper %T Learning Efficient Guardrails for Compliance %A Xiaofei Wen %A Wenjie Jacky Mo %A Yanan Xie %A Peng Qi %A Muhao Chen %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-wen26g %I PMLR %P 134614--134630 %U https://proceedings.mlr.press/v306/wen26g.html %V 306 %X Autonomous web agents are increasingly deployed for long-horizon tasks, yet their ability to adhere to real-world policies remains critically underexplored compared to standard safety objectives. To address this gap, we introduce PolicyGuardBench, a benchmark of 60k policy-trajectory pairs designed to evaluate compliance through both full-trajectory and novel prefix-based violation detection tasks. Using this dataset, we train PolicyGuard, a lightweight guardrail model that achieves strong detection accuracy while maintaining high inference efficiency. Notably, our model demonstrates robust generalization capabilities, preserving high performance even on unseen domains. These contributions establish a comprehensive framework for studying policy compliance, showing that accurate and generalizable guardrails are feasible at small scales.
APA
Wen, X., Mo, W.J., Xie, Y., Qi, P. & Chen, M.. (2026). Learning Efficient Guardrails for Compliance. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:134614-134630 Available from https://proceedings.mlr.press/v306/wen26g.html.

Related Material