VERA: A Responsible AI Dashboard for Assessing LLM Deployments in Light of the EU AI Act

VERA translates complex EU AI Act requirements into measurable technical standards: The dashboard operationalizes the EU AI Act through COMPL-AI's framework of six principle groups and eighteen technical requirements, making regulatory compliance concrete and assessable for non-technical decision-makers approving AI deployments.

The dashboard cuts assessment time nearly in half: In a within-subject study, reviewers completed responsible AI assessments in 30 seconds using VERA compared to 56 seconds using raw benchmark output files—a 46% reduction that demonstrates how better interface design can streamline compliance workflows.

Environmental impact is now part of the compliance picture: VERA instruments energy consumption during model evaluations, adding quantifiable evidence about the environmental footprint of LLM deployments alongside traditional responsible AI metrics—addressing a dimension often overlooked in AI governance.

Current benchmarks can measure 12 of 18 EU AI Act requirements: While VERA successfully scores twelve measurable requirements, the paper identifies a gap: six requirements lack current benchmark coverage, indicating where the responsible AI measurement field still needs development.

Non-technical staff found the dashboard highly useful for governance: With a usefulness rating of 4.36 out of 5 from organizational reviewers, VERA demonstrates that translating technical evidence into accessible dashboards can effectively bridge the gap between AI engineers and the non-technical stakeholders responsible for deployment approval decisions.