Atlas of AI Risk

A field guide to the full risk surface of advanced AI — from misuse to misalignment to slow structural drift.

Summary

You cannot govern a fog. So this volume draws a map. Every serious conversation about AI risk currently suffers the same failure: people use one word, "risk", for a dozen different things, then wonder why they disagree. This atlas maps the full surface as four continents: Misuse, Malfunction, Misalignment, and Structural drift, divided into twelve territories, each scored for severity and tractability. It shows how risks compound across borders, and where to build the watchtowers that would see trouble coming. A shared map does not settle the argument. It makes the argument productive.

Volume 03 is the series' field guide to what can actually go wrong. It maps the full risk surface of advanced AI as four continents (Misuse, Malfunction, Misalignment, and Structural drift) divided into twelve territories, each scored for severity and tractability. It traces the trade winds by which risks compound and convert across continents, and closes with watchtowers: observable early-warning indicators, a recommendations agenda, and a working glossary. The atlas's premise is simple: confusion about the map is itself a risk.

Key findings

1. Taxonomy confusion is itself a risk Attention follows vividness, budgets follow attention, and exposure accumulates wherever the categories are blurry. The first duty of risk management is a map everyone can point at, including to disagree.
2. Four continents, four kinds of physics Misuse is deliberate harm by people; Malfunction is systems failing at their jobs; Misalignment is systems pursuing the wrong thing well; Structural drift is harm nobody chose. Each responds to different instruments, which is the point of separating them.
3. The weights sit in different decades Near-term expected harm concentrates in Misuse and Malfunction; the heaviest tails live in Misalignment; the most certain losses are Structural, because drift is what happens by default. Portfolio design must hold all three horizons at once.‐
4. Risks trade across borders A misuse incident taxes trust everywhere; chronic malfunction converts to institutional distrust; racing dynamics thin alignment margins. Because risks compound and convert, single-risk policies systematically underperform portfolios.
5. Tractability varies more than severity The atlas's most actionable fact: several severe territories are quite tractable (reliability standards, fraud defenses, fallback requirements) while others buy little with money alone. Spend where a decade of effort moves the map.
6. Early warning is buildable, and mostly unbuilt Every continent has observable leading indicators, evaluation gaps, fallback coverage, concentration indices, junior-pipeline health. Chapter 07 names nine watchtowers; most currently have no keeper.

How to cite this report

Atlas of AI Risk: A field guide to the full risk surface of advanced AI — from misuse to misalignment to the slow structural drift. FAIR Labs Frontier Series, Vol. 03. Fair Artificial Intelligence Research Labs, 2026. Available at https://fairlabs.ai/research/frontier-vol03-atlas-of-ai-risk

Topics: AI risk taxonomy, misuse, misalignment, systemic risk, early warning, risk assessment

FAIR Labs (Fair Artificial Intelligence Research Labs) is a nonpartisan 501(c)(3) research institute in Washington, DC. Its reports may be reproduced with attribution for non-commercial purposes. For interviews, briefings or data requests, use the contact form.

Related FAIR Labs reports