Capable but Careless: Do Computer-Use Agents Follow Contextual Integrity?

📰 ArXiv cs.AI

Learn how to evaluate computer-use agents' ability to follow contextual integrity and mitigate privacy risks

advanced Published 23 Jun 2026

Action Steps

Build an evaluation harness like AgentCIBench to test agents' contextual integrity
Run experiments to assess agents' ability to follow contextual rules
Configure agents to access multiple applications and evaluate their information-sharing behavior
Test agents' performance in different contexts to identify potential privacy risks
Apply evaluation results to improve agents' design and mitigate privacy risks

Who Needs to Know This

Developers and researchers working on computer-use agents and privacy preservation can benefit from understanding how to evaluate agents' contextual integrity

Key Insight

💡 Computer-use agents can pose privacy risks if they pull in information from one context to another inappropriately

Key Takeaways

Learn how to evaluate computer-use agents' ability to follow contextual integrity and mitigate privacy risks

Full Article

Title: Capable but Careless: Do Computer-Use Agents Follow Contextual Integrity?

Abstract:
arXiv:2606.23189v1 Announce Type: new Abstract: Computer-use agents (CUAs) now act on a user's behalf across personal applications such as email, calendars, and to-do lists. This cross-application access is useful, but it also creates a privacy risk that has been largely overlooked: when an agent works in one context, it can pull in information from another that is inappropriate in that context. Hence, we introduce AgentCIBench, an evaluation harness that turns this risk into executable, determi

Read full paper → ← Back to Reads