Public beta

AI-Risk.ai is live and evolving. Verified evidence is shown where available; deeper coverage, history, watchlists and alerts are being added.

See what's live →
Transparent by design

Methodology

AI-Risk.ai separates evidence from interpretation. We track what changed, explain why it may matter, show who could be exposed, state confidence, and identify the next signal that could change the picture.

AI Impact Monitor

Signal first. Score only when justified.

We do not publish a synthetic all-purpose AI risk number simply because one looks good in a dashboard. Each dimension becomes quantitative only when source quality, update cadence and historical comparability support it.

01 · What changed

Detect meaningful movement

Compare current evidence with a relevant baseline or prior release instead of treating every headline as a signal.

02 · Why it matters

Translate evidence into context

Explain the transmission path: how a change in AI capability, adoption, labor markets, financing or geopolitics could affect real decisions.

03 · Watch next

Make uncertainty useful

State confidence and name the next data point that would confirm, weaken or materially change the current interpretation.

First deep-dive

AI Jobs & Work Monitor

The U.S.-first labor-market monitor deliberately separates technical exposure from observed AI use and realized labor-market outcomes.

Structural exposure

What can AI do?

Occupation tasks and skills are anchored in O*NET. External task-level exposure research, including the AI Risk Index, can be used as one versioned input with attribution.

Observed adoption

What is actually being used?

Anthropic Economic Index, Census BTOS and other published adoption evidence help distinguish technical capability from real deployment.

Labor-market outcomes

What is changing in work?

BLS, Stanford and future labor-market feeds track hiring, entry-level access, wages and skill demand. A high exposure score alone is never treated as a forecast of unemployment.

External providers are inputs, not runtime dependencies. Versioned provider data is designed to be stored in our own database before product use, so an upstream outage cannot silently rewrite or break the user experience.
Context layer · model v1.1

Global Risk Pulse

Global Risk Pulse is a separate macro context model. Version 1.1 keeps the original 30/40/30 weighting while replacing two fragile news-coverage proxies with dedicated, reproducible research datasets.

Energy · 30%

Brent price movement

Daily Brent Europe spot-price observations are retrieved from FRED, sourced from the U.S. Energy Information Administration. The latest daily percentage move is translated into the energy stress signal.

Source: FRED / EIA
Conflict · 40%

AI-GPR geopolitical pressure

The daily AI-GPR Index by Matteo Iacoviello and Jonathan Tong measures geopolitical risk using LLM-scored newspaper coverage. AI-Risk.ai takes the current 7-day average and converts its position within the trailing five-year distribution into a 0–100 pressure score.

This is a geopolitical-risk intensity measure, not a probability of war and not independent event verification.

Source: Iacoviello & Tong AI-GPR
Supply · 30%

NY Fed supply-chain pressure

The New York Fed Global Supply Chain Pressure Index combines transportation costs and manufacturing indicators into a monthly measure expressed in standard deviations from its historical average.

For the 0–100 context scale, zero standard deviations maps to 50 and each standard deviation shifts the signal by 20 points, capped at 0 and 100.

Source: Federal Reserve Bank of New York GSCPI

Global Risk Pulse = Energy × 30% + Conflict × 40% + Supply × 30%

A composite is shown only when all three required inputs are backed by real observations. Source mode, provenance, model version and leading drivers remain visible; a source outage never becomes a fabricated live value.

AI-assisted summaries and interpretations do not guarantee future events. AI-Risk.ai is an informational intelligence product, not a prediction engine, person score, employment verdict or investment recommendation. Uncertainty and conflicting evidence are part of the output, not something to hide.