TRUST-CUA: Trustworthy Computer-Using Generalist Agents for Intelligent User Interfaces

Limassol, Cyprus

Jul 13, 2026

Welcome

TRUST-CUA @ IUI 2026 brings together the IUI and AI-agent communities to explore how we can build predictable, steerable, and most importantly, trustworthy computer-using agents (CUAs). These agents operate across GUIs, browsers, APIs, and CLIs, and increasingly act as multi-modal collaborators that assist users in completing complex digital tasks.

Workshop Agenda

09:00 – 09:15 | Welcome & Opening Remarks

Segev Shlomov (IBM Research)

Welcome to TRUST-CUA @ IUI 2026
Workshop goals and motivation
Beyond Demos: From impressive prototypes to reliable, trustworthy computer-using agents
Overview of the day's program

09:15 – 09:40 | Invited Talk

Rotem Dror (University of Haifa)

Data Annotations in the Era of Large Language Models

09:40 – 10:30 | Technical Session I — Trust, Evaluation & Safety

Beyond Task Completion: A Process-Oriented Framework for Evaluating Preference and Value Alignments in Computer-Using Agents
Remote Presentation

Language Model Agents Under Attack: A Cross-Model Benchmark of Profit-Seeking Behaviors in Customer Service
Pre-recorded Presentation

Trust by Design: Trust Calibration Through Non-Advisory Socratic Dialogue in Conversational Agents
Pre-recorded Presentation

10:30 – 11:00 | Coffee Break ☕

11:00 – 12:10 | Technical Session II — Reliable Architectures & Human-Centered Agent Design

Architecting Trust Without Killing Autonomy: Borrowing Microservice Resilience Patterns for Reliable Computer-Using Agents
Remote Presentation

Designing Trustworthy Agentic RAG Systems Through Modality-Separated Interaction and Error-Aware Memory
Remote Presentation

Zero: Towards Reliable Agentic UX for Industrial Dashboards
Remote Presentation

AgentFixer: From Failure Detection to Fix Recommendations in LLM Agentic Systems
Pre-recorded Presentation

12:10 – 12:30 | Open Discussion & Closing

An interactive discussion on the future of trustworthy computer-using agents, including:

Reliability and robustness
Human oversight and trust
Evaluation beyond task completion
Future benchmarks and community collaboration

Closing remarks and next steps.

12:30 – 13:30 | Lunch 🍽️

Workshop Focus

Modern computer-using agents are evolving into generalist, multi-agent systems capable of reasoning across diverse tools and interfaces. They hold tremendous promise for improving productivity, creativity, and automation. However, they also introduce new challenges for reliability, transparency, and user oversight, and despite this potential, real-world business adoption remains limited.

Developing and validating such systems is often slow and costly, especially when safety and compliance are at stake. Even after deployment, ensuring reliability and trustworthiness is difficult: agents can make silent mistakes, repeat past errors, or drift from intended behavior without clear user visibility.

TRUST-CUA @ IUI 2026 addresses these challenges by exploring how to design CUAs that are predictable, auditable, and user-trustable. The workshop focuses on interface and UX paradigms that foster trust, explainability, human-in-the-loop (HITL), and control, as well as evaluation and governance frameworks that ensure accountability in both enterprise and public settings.

By connecting AI, UX, and human-centered design, TRUST-CUA aims to define a roadmap toward reliable, transparent, and production-ready generalist agents that operate safely and effectively across domains.

Call for Papers