The Digital Press

Modern AI systems evolve quickly and can cause serious harm, but current risk assessments often miss key pathways and assumptions. This paper introduces a probabilistic risk assessment (PRA) framework adapted from high-risk industries to systematically evaluate AI risks with clear, evidence-based methods.

Why it matters: The PRA framework offers a rigorous way to identify and quantify AI risks, improving safety and decision-making.

The big picture: It adapts proven risk tools from nuclear and aerospace industries to address AI's unique hazards and uncertainties.

Stunning stat: The framework integrates multiple risk pathways into aggregated, quantified absolute risk estimates for critical decisions.

Quick takeaway: A publicly available workbook guides AI developers and regulators through systematic, transparent AI risk assessment.