LEASH: Uncertainty-Aware Weak-to-Strong Escalation and Oversight for Computer-Use Agents

Introduces LEASH, a training-free weak-to-strong escalation framework for computer-use agents that uses internal hidden state uncertainty to dynamically trigger handoff to a strong model and supervise it via a shadow observer on OSWorld.

• OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use
  October 15, 2026 · arXiv
• OS-Helix: Unifying GUI Interaction and Code Execution in Native Computer-Using Agents
  October 15, 2026 · arXiv
• VeriCUA: Learning Proactive Verification for Reliable GUI Agent Rewards
  October 19, 2026 · arXiv
• CUA-Lite: An Open Platform for Computer Use Agents
  October 19, 2026 · arXiv
• Agent World Model in the Wild: Computer-Use Agent Environment Synthesis
  October 19, 2026 · arXiv
• Adaptation of Computer Use Agents to New Software with User Manuals
  October 16, 2026 · arXiv
• AgentHorizon: Evaluating Agentic Judges for Long-Horizon Computer-Use Tasks
  October 8, 2026 · arXiv
• RSIAgent: Autonomous Exploration for Recursive Self-improvement in New Environments
  September 14, 2026 · arXiv
• LiteCUA: Computer as MCP Server for Computer-Use Agent on AIOS
  May 24, 2025 · arXiv
• Agent S2: A Compositional Generalist-Specialist Framework for Computer Use Agents
  April 1, 2025 · COLM 2025
• Evolving Agent Interfaces
  October 20, 2026 · arXiv
• Comparing Computer Use Behavior Across Multiple Levels of Abstraction
  October 19, 2026 · arXiv
• PersonaCUABench: Benchmarking Computer-Use Agents Serving Diverse Personas
  October 16, 2026 · arXiv
• Task-State Governance: A Virtual Machine for Verifiable and Recoverable GUI Agent Execution
  October 14, 2026 · arXiv
• MAI-Gym: An Open All-in-One Platform for GUI Agent Evaluation and RL Training across Mobile and Computer Environments
  October 14, 2026 · arXiv
• Pack Light, CarryOn: Building Object-Oriented Context for Long-Horizon Agents
  October 13, 2026 · arXiv
• Learning How to Remember: Memory Modality Selection for GUI Agents
  October 13, 2026 · arXiv
• A Decade in a Day: Synthesizing Agent-Native Interfaces from Human-Oriented Software Evolution
  October 11, 2026 · arXiv
• Work2Skill: Capturing Human GUI Experience for Reusable Agent Skill Induction
  October 9, 2026 · arXiv
• Latent World Action Model for GUI Agents
  October 9, 2026 · arXiv
• Control OSWorld: An AI Control Environment for GUI Computer Use Agents
  October 2, 2026 · arXiv
• CUAWright: A Minimal Unified Interface for Digital Agents
  October 2, 2026 · arXiv
• GUI-HARVEST: Self-Improving GUI Agents through Evidence-Driven Harness Evolution
  October 1, 2026 · arXiv
• AutoGUIWorld: Image Generators as Visual World Models for GUI Agent
  October 1, 2026 · arXiv