Just Do It!? Computer-Use Agents Exhibit Blind Goal-Directedness
Erfan Shayegani , Keegan Hines , Yue Dong , Nael Abu-Ghazaleh , Roman Lutz , Spencer Whitehead , Vidhisha Balachandran , Besmira Nushi , Vibhav Vineet
- 🏛 Institutions
- Microsoft Research AI Frontiers , Microsoft AI Red Team , University of California , Riverside , NVIDIA
- 📅 Date
- October 2, 2025
- 📑 Publisher
- ICLR 2026 (Poster)
- 💻 Env
- General GUI
- 🔑 Keywords
TLDR
This paper identifies Blind Goal-Directedness (BGD), where computer-use agents continue pursuing goals despite feasibility, safety, reliability, or context concerns. It introduces BLIND-ACT, a 90-task benchmark on OSWorld, and finds high average BGD rates across frontier models even after prompting-based mitigations.
Related papers (24)
- Agent World Model in the Wild: Computer-Use Agent Environment SynthesisOctober 19, 2026 · arXiv
- AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use AgentsApril 3, 2026 · arXiv
- Agent Alpha: Tree Search Unifying Generation, Exploration and Evaluation for Computer-Use AgentsFebruary 3, 2026 · arXiv
- Agentic Reward Modeling: Verifying GUI Agent via Online Proactive InteractionJanuary 31, 2026 · arXiv
- BEAP-Agent: Backtrackable Execution and Adaptive Planning for GUI AgentsJanuary 29, 2026 · arXiv
- From Off-Policy to On-Policy: Enhancing GUI Agents via Bi-level Expert-to-Policy AssimilationJanuary 9, 2026 · arXiv
- R-WoM: Retrieval-augmented World Model For Computer-use AgentsOctober 13, 2025 · ICLR 2026 (Poster)
- Agent S2: A Compositional Generalist-Specialist Framework for Computer Use AgentsApril 1, 2025 · COLM 2025
- Evolving Agent InterfacesOctober 20, 2026 · arXiv
- CUA-Lite: An Open Platform for Computer Use AgentsOctober 19, 2026 · arXiv
- OpenGCUA: Toward a Fully Open Stack for Generalist Computer UseOctober 15, 2026 · arXiv
- OS-Helix: Unifying GUI Interaction and Code Execution in Native Computer-Using AgentsOctober 15, 2026 · arXiv
- LEASH: Uncertainty-Aware Weak-to-Strong Escalation and Oversight for Computer-Use AgentsOctober 15, 2026 · arXiv
- MAI-Gym: An Open All-in-One Platform for GUI Agent Evaluation and RL Training across Mobile and Computer EnvironmentsOctober 14, 2026 · arXiv
- Pack Light, CarryOn: Building Object-Oriented Context for Long-Horizon AgentsOctober 13, 2026 · arXiv
- Learning How to Remember: Memory Modality Selection for GUI AgentsOctober 13, 2026 · arXiv
- Work2Skill: Capturing Human GUI Experience for Reusable Agent Skill InductionOctober 9, 2026 · arXiv
- Latent World Action Model for GUI AgentsOctober 9, 2026 · arXiv
- AgentHorizon: Evaluating Agentic Judges for Long-Horizon Computer-Use TasksOctober 8, 2026 · arXiv
- Control OSWorld: An AI Control Environment for GUI Computer Use AgentsOctober 2, 2026 · arXiv
- CUAWright: A Minimal Unified Interface for Digital AgentsOctober 2, 2026 · arXiv
- GUI-HARVEST: Self-Improving GUI Agents through Evidence-Driven Harness EvolutionOctober 1, 2026 · arXiv
- AutoGUIWorld: Image Generators as Visual World Models for GUI AgentOctober 1, 2026 · arXiv
- Talk2Agent: Benchmarking Voice Interfaces for Text AgentsSeptember 30, 2026 · arXiv