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

Paphos, Cyprus

March 23-26, 2026

Welcome

TRUST-CUA: brings the IUI community together around a pressing question: how do we design predictable, steerable, auditable, and safe computer-using agents (CUAs) that operate across GUIs, browsers, APIs, and CLIs.

All workshop materials (program, slides, papers) will be posted here once available.

Agenda

14:00 – 14:10 Gathering
14:10 – 14:30 Introduction & Opening Remarks (Organizers)
14:30 – 15:15 Keynote: Human-Centered Trust in Agentic UIs (speaker TBA)
15:15 – 15:30 Paper Spotlight I
15:30 – 16:00 Coffee Break
16:00 – 16:30 Paper Spotlight II
16:30 – 16:40 Lightning: Benchmarks & Shared Tasks for Trustworthy CUAs
16:40 – 17:10 Tutorial/Demo: Oversight & Recovery UX for CUAs
17:10 – 17:30 Brainstorm & Conclusions: Toward the TRUST-CUA Checklist & CUBench-IUI

Topics

Topics include (but are not limited to):

Safe & Trustworthy Agents: staged execution; detect–explain–recover loops; dashboards; rollback/undo; policy-adherent behavior; reliability over long horizons.
Human–Agent Interaction: mixed-initiative control; pause/approve checkpoints; interactive debugging; trace/provenance visualization; uncertainty & risk communication.
Guardrails & Governance: least-privilege consent; policy visualization; norm-aware affordances; interfaces for audit, compliance, and red-teaming.
Self-Evolving Agents: safe learning from feedback; controlled prompt/config tuning under UX/policy constraints; avoiding error repetition & drift.
Knowledge & Learning: documents/trajectories/demos as customization; preference elicitation; modular adaptation surfaced through controllable UX.
Evaluation & Benchmarks: user-centered metrics (predictability, oversight burden, time-to-recovery); reproducibility artifacts; datasets and protocols.
Real-World Evidence: deployments, incidents & near-misses, anti-patterns; organizational and enterprise contexts.
Foundations & Methods: planning & decision-making for controllability and explanation; program synthesis/validation; HCI methods for agentic UX.