VeriCUA: Verification as Agentic Exploration for Computer-Using Agents
Qiushi Sun, Da Yu, Ravi Kumar, Pritish Kamath and colleagues (Google and the University of Hong Kong) present VeriCUA, a platform-agnostic framework that verifies computer-using agents by running verification as an agentic exploration task.
Ask this paper
Method. The verifier inspects the available context, builds a task rubric, selects GUI or code checks for unresolved requirements, and combines findings under explicit evidence rules.
False positives. Across three online environments, VLM judges pass 15.4% to 27.9% of failed runs. VeriCUA lowers that to 5.4% to 13.9%.
Accuracy. Online accuracy rises from 71.6% to 78.4% with the VLM judge to 79.5% to 90.5% with VeriCUA, with abstentions counted as failures.
Abstention. VeriCUA decides 70% to 82% of runs. Accuracy on decided runs is 88.7% to 95.1%.
Offline. It also has the highest offline accuracy across the compared benchmarks, with the largest advantage on hard trajectories.
Abstract
Building and scaling computer-using agents (CUAs) across desktop, mobile, and web environments requires reliable task-completion verification for data curation, evaluation, and reinforcement learning. Typical automated approaches prompt vision-language models (VLMs) to judge trajectories or generate programs to inspect application state. However, both rely on partial views of complex digital environments. VLM judges are restricted to supplied observations, while generated programs depend on accessible state. These views can omit decisive evidence or leave completion judgments unsupported. Tailoring separate verifiers to different platforms and access settings adds further engineering cost. We present VeriCUA, a platform-agnostic framework for verification through agentic exploration. It treats verification as dynamic evidence gathering and assessment within a shared procedure for online and offline settings across platforms and observation modalities. After inspecting the available context to construct a task rubric, the verifier selects GUI or code checks for unresolved requirements and combines their findings under explicit evidence rules. Experiments show improved agreement with reference labels over the compared baselines across these settings, with the largest offline gains on hard cases and stronger rejection of false completion claims online. Further analyses connect these gains to complementary evidence and test-time exploration.