EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Shielded Agentic RL with Formal Failure-Set Constraints for Safe Web Actions under Budget Limits

Daniel Thompson, Emily Chen, Michael R. Brown

American Journal of Machine Learning · 25.03.2026 · 7 (1) · 110-117

DOI: 10.71465/ajml3652 ↗

Kurz gefasst

Fachartikel in der Zeitschrift American Journal of Machine Learning, online erschienen am 25.03.2026. Verfasst von Daniel Thompson und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Enterprise web agents face safety-critical constraints where certain action sequences can cause irreversible harm. We propose a shielded agentic RL framestudy that integrates a lightweight action shield with constrained policy learning. The shield uses (i) a learned classifier over DOM/state features to detect proximity to a predefined failure set (e.g., destructive actions, payment confirmation), and (ii) a rule+model hybrid verifier that blocks or rewrites actions when predicted failure probability exceeds a budget-aware threshold. The underlying policy is optimized via constrained RL to minimize multi-cost usage while maximizing task success, with the shield providing safety guarantees during both training and deployment. We recommend evaluation on safety-focused benchmarks (e.g., enterprise web agent safety suites) with 500–1,500 tasks, measuring prevented unsafe actions, residual failure rate, added latency overhead, and success under strict cost caps. The proposed design emphasizes auditability: the shield logs blocked actions and provides human-readable rationales, enabling compliance-oriented deployment.

Zitieren

APA

Thompson, D., Chen, E., & Brown, M. R. (2026). Shielded Agentic RL with Formal Failure-Set Constraints for Safe Web Actions under Budget Limits. American Journal of Machine Learning, 7(1), 110-117. https://doi.org/10.71465/ajml3652

Vancouver

Thompson D, Chen E, Brown MR. Shielded Agentic RL with Formal Failure-Set Constraints for Safe Web Actions under Budget Limits. American Journal of Machine Learning. 2026;7(1):110-117. doi:10.71465/ajml3652

BibTeX
@article{Thompson2026,
  author = {Thompson, Daniel and Chen, Emily and Brown, Michael R.},
  title = {Shielded Agentic RL with Formal Failure-Set Constraints for Safe Web Actions under Budget Limits},
  journal = {American Journal of Machine Learning},
  year = {2026},
  volume = {7},
  number = {1},
  pages = {110--117},
  doi = {10.71465/ajml3652}
}
Kostenlos starten →