CORA: Conformal Risk-Controlled Agents for Safeguarded Mobile GUI Automation

Yushi Feng , Junye Du , Qifan Wang , Zizhan Ma , Qian Niu , Yutaka Matsuo , Long Feng , Lequan Yu

🏛 Institutions: HKU , CUHK , University of Tokyo
📅 Date: April 10, 2026
📑 Publisher: arXiv
💻 Env: Mobile
🔑 Keywords: safety benchmark conformal risk control Phone-Harm CORA

TLDR

CORA reformulates safety as selective action execution: a Guardian model estimates action-conditional risk, Conformal Risk Control calibrates an execute/abstain boundary under a user-specified risk budget, and a Diagnostician proposes interventions for rejected actions. A Goal-Lock mechanism resists visual injection. Phone-Harm benchmark with step-level harm labels is also released.

Open paper arXiv Report issue