Towards Safe Autonomy for Computer Use

Summary

SA-CUA is a dual-LLM framework for Computer Use Agents, which act visually on screens and lack a boundary between trusted and untrusted content. It derives conservative deterministic rules from the accessibility tree to unmask trusted regions from the privileged LLM's screenshot, and uses an endorsement channel to verify untrusted values before they reach the trusted domain. The authors report it keeps much higher utility than CaMeL-CUA while holding the attack success rate at zero.

Topics

• AI agents

Related items

• MediumVisual Studio Copilot and SSMS Remote Code Execution (CVE-2026-103572)Similar attack · Embrace The Red
• HighCVE-2026-110206: Roo Code command approval bypass in parseCommandLine() via command substitutionSimilar attack · NVD/CVE Database
• HighCVE-2026-110205: Roo Code command approval bypass through parseCommandLine() command substitutionSimilar attack · NVD/CVE Database
• HighCVE-2026-110204: Roo Code command approval bypass through parseCommandLine substitutionsSimilar attack · NVD/CVE Database
• HighCVE-2026-110203: Roo Code command approval bypass via quote-masking regexSimilar attack · NVD/CVE Database