How Warning Shots works

The rules behind the scores, the Wire, the forecasts and the fact-checking, in one place.

What counts as an incident

Real-world harms or near-misses caused or materially enabled by an AI system, and lab or evaluation findings of dangerous behavior. This is a curated index of significant, well-sourced incidents, not a census; the AI Incident Database and the OECD monitor count far more.

How incidents are scored

Each incident is rated on five factors, then scored by a fixed formula.

Harm (H, 0–5)

Damage actually done: none, embarrassment, material loss, a death or $1M+, multiple deaths or $100M+ or infrastructure, mass casualty or $1B+.

Reach (R, 0–5)

Who was affected: a lab, one person or organization, dozens, thousands, millions, hundreds of millions.

Autonomy (A, 0–4)

Tool (a person misused it), relied on (people trusted bad output), acted (a bounded action), autonomous (multi-step operations on its own), against intent (worked against its own operator or developer).

Novelty (N, 1–5)

Repeat of a known pattern, escalation, first real-world case of a lab behavior, new dangerous capability, paradigm shift.

Irreversibility (I, 0–5)

Caught in a sandbox, easily undone, undone with effort, partly permanent, permanent, permanent and spreading.

Harm score = 100 × (0.5·H/5 + 0.25·R/5 + 0.25·I/5)
Signal score = 100 × (0.6·A/4 + 0.4·N/5)
Severity = 0.55 × Harm + 0.45 × Signal

Tiers: Minor under 20, Moderate 20–39, Serious 40–59, Severe 60–79, Critical 80 and up, shown as WS-1 to WS-5. Harm is weighted more because it already happened; signal is weighted heavily because it predicts what happens next. A lab finding can reach Serious on signal alone, but only real-world damage can make an incident Severe or Critical. The rating is an editorial judgment by a published formula, not a probability; evidence confidence is recorded separately. Each incident has a stable dispatch number (WS-0001 onward, in date order).

The proof ladder: Tried, Proven, Hit

For a live attack, a rung is climbed only on a named, linked source. Tried: attacks were attempted and logged; nothing is confirmed broken into. Proven: a test or decoy system was actually broken into; no real victim confirmed. Hit: a real organization confirmed it was breached. The three are never blurred in a headline.

The Wire

A board of links to other people's reporting. Each headline is written by Warning Shots in its own words, names the outlet, links straight to it and stays on the board for about three days. Every story is checked against the original page before it is added; stories that cannot be opened are left off. A line may sharpen and pick the most alarming true fact, but never claim more than the story supports: no “hacked” for a test, no “victims” for decoys, no figures rounded up. When only reporters, not the company, tie a story to AI, it is labelled Reported.

Forecasts

Each forecast is a dated yes-or-no question with the index's odds, published before the answer is known. The wording never changes after it goes up; each forecast page shows a fingerprint (SHA-256) of the exact question and the commit that added it. Calls are graded in public on the day they resolve, misses stay on the board, and a call at exactly 50% is graded but not counted. Odds can move before resolution only on evidence, and every move is logged.

Fact-checking

Every fact is checked twice. First as it is written, from a page opened that day. Then in a separate pass before publishing: each headline and line is re-checked against its source word by word, every computed number is re-derived from the data, and the pass looks for duplicates, things counted as events that aren't (surveys, industry tallies, tests), absolute words (“every”, “first”, “none”) and any number that differs between two places on the site. The build refuses to publish when a headline's figure is missing from its own story or a key figure differs from the data.

Images and AI

Photos are free-licensed (Unsplash, Pexels, Wikimedia) and credited; they illustrate, and never show the real people or places. No AI-generated images of real incidents. The site is written with AI assistance and checked against its sources.

Corrections → · About →

Spot an error or something out of date? Tell us; corrections are made and logged. Written with AI assistance from public sources; checked against those sources before publishing. General information, not legal, medical or financial advice.