Systemic risk pipeline and dashboard reveal ai safety weak spots
An Open Pipeline and Dashboard for Systemic-Risk Evidence under the EU AI Act's Code of Practice
Artificial Intelligence
Summary
AI safety claims often lack clear or easy-to-see evidence. The authors created a public tool called the Systemic Risk Index that organizes many safety tests into meaningful categories and shows clear scores for AI models. This tool lets people see how risky an AI model might be on average or in the worst cases, making hidden risks easier to find. They also showed their approach works well by comparing AI and human judgments and testing if safety-related changes really keep harms the same.
What this means in practice
- •For ai product teams: Evaluate AI model risks transparently using an interactive dashboard that highlights worst-case and average harms across multiple safety categories.
- •For ai policy compliance teams: Support alignment with EU AI Act safety guidelines by tracing risk assessments back to specific benchmarks and providing configurable risk scores.
Authors
Jacob T. Emmerson, Phuong-Anh Nguyen-Le, Ronan Romano, Wilber Sean V. Anterola, Yann Billeter, Zhijing Jin
Abstract
Claims about AI safety reach audiences well beyond the AI community, yet many rely on opaque evidence or static assessments, when supporting evidence is accessible at all. We present the Systemic Risk Index, an open evaluation pipeline and dashboard built to make empirical evidence more transparent and traceable to the public. Our work organizes 19 public benchmarks into four systemic-risk categories defined by the EU GPAI Code of Practice---CBRN, cyber offense, harmful manipulation, and loss of control---and evaluates models using harm-preserving perturbations and simulated deployment contexts. The interactive dashboard lets users alternate between average and worst-case aggregation, vary how model capability affects the aggregate score, and trace each risk rating to its benchmark evidence. Across 18 models, scores fall by 14 to 37 points under worst-case aggregation, highlighting information that can be hidden by an average assessment of model risk. LLM judges show agreement with human graders comparable to human--human agreement ($κ= 0.78\text{--}0.82$), and a blind audit finds that $83\%$ of sampled transformations preserve the original harm. In a survey ($N = 21$), most participants report that scores are easy to understand and that the dashboard encouraged them to view model evaluations under different settings