Expert Prioritization Survey

A comprehensive expert assessment of which AI risks matter most, who is vulnerable, and who should be responsible.

What was the Expert Prioritization Survey?

The MIT AI Risk Expert Prioritization Survey engaged 272 AI experts from around the world to identify which AI risks are most important, who is most affected, and who should take action. The survey covered all 24 domains in the MIT AI Risk Repository, 7 AI ecosystem actors, and 14 industry sectors. Read on for more of the details.

What can I use this for?

Use this data to understand which AI risks experts consider most urgent, identify which sectors and stakeholders face the greatest exposure, and see where responsibility gaps exist between those who are vulnerable and those who should act. Pair with the governance and incident datasets to compare expert priorities against regulatory coverage and real-world harms.

18 of 24

risk domains have ≥10% catastrophic probability within 5 years

information, finance & national security

are the most vulnerable sectors

AI users and affected stakeholders

bear most vulnerability

developers, governments, and regulators

are assigned most responsibility

Severity Distributions

Joy plot of expert severity distributions across 24 subdomains, comparing Business as Usual vs Pragmatic Mitigations with catastrophic risk annotations.

Each ridge shows the probability distribution across five severity levels (negligible to catastrophic) as assessed by experts. Rows are sorted by BAU severity (highest on top). The right column shows the percentage of experts who assigned “catastrophic” severity under each scenario. Click any row for detailed IQR breakdown.

Interactive Visualizations

Click on any visualization to explore expert consensus data in detail.

Concern Rankings

What risks do experts worry about most? Ranked by top-3 concern selections.

Severity × Likelihood Scatter

BAU and mitigated severity vs likelihood with 3-way toggle.

Severity × Likelihood Heatmap

Severity probability distributions per subdomain with IQR drill-down.

Severity Distributions

Joy plot of severity distributions comparing BAU vs Pragmatic Mitigations.

Severity Rankings: BAU vs Mitigated

Per-subdomain severity distributions for BAU vs Pragmatic Mitigations, ranked by mitigation gap.

Actor × Risk Responsibility

How responsible is each actor type for each risk? Heatmap with consensus filtering.

Actor × Risk Vulnerability

How vulnerable is each actor type to each risk? Heatmap with consensus filtering.

Vulnerability vs Responsibility

Scatter comparing vulnerability and responsibility for a selected actor type.

Actor Vulnerability vs Responsibility

For each actor, how do responsibility and vulnerability ratings compare across subdomains?

Actor × Subdomain Vuln/Resp Matrix

How do experts rate every actor's responsibility and vulnerability across every subdomain at once?

Moral Hazard Scatter

Which actors are vulnerable to AI risks but not responsible for managing them?

Sector × Subdomain Matrix

How vulnerable is each of 14 sectors to each risk? Heatmap with consensus filtering.

Single-Sector Profile

All subdomains ranked by vulnerability for a selected sector.

Data sourced from the Delphi expert elicitation study. Severity and likelihood ratings reflect expert consensus under Business as Usual (BAU) and Pragmatic Mitigations (PM) scenarios. Actor and sector ratings use weighted means of 5-point Likert distributions.