Severity & Concern

Each point represents a specific AI risk, positioned by its expected severity (x-axis) and likelihood of significant harm (y-axis) based on expert assessments. Points further up and to the right indicate risks that are both more severe and more likely. Filled vs. outlined markers show baseline vs. mitigated scenarios, with connecting lines illustrating how interventions could shift risk levels.

Each row shows a risk, and each column shows the percentage of expert responses falling into different severity levels (from negligible to catastrophic). The color intensity reflects concentration. The “Avg” column provides a summary score. Read across a row to understand how experts distribute their expectations.

This chart ranks AI risks by overall level of expert concern. Longer bars indicate risks that a larger share of experts identified as important or high-priority. Use this view to quickly see which risks rise to the top across the panel and compare their relative prominence.

Each row is a risk subdomain. Bars show how experts rate the risk under Business as Usual — by default the percentage who chose Catastrophic, sorted from highest to lowest. Use View mode to switch the bars to a different severity level, the weighted-mean severity (1–5 scale), or the full distribution stack. Toggle Show mitigations to overlay Pragmatic Mitigations alongside BAU; a right-hand Change column then reports the PM − BAU shift, green when mitigations push severity in the helpful direction and red when they don’t. Switch Data view to cluster subdomains under their parent domains or show domain averages only, and click any subdomain row to open its full severity distribution.