How the numbers work

Every number on this site comes from a public record and links to it. The data and the code that produce it are public, and the site updates itself through the day.

The Fear Index

Each of the nine fears is scored from 0 to 100 against the loudest fear on each channel. Legislation counts 40%: bills, rules and orders citing the fear since January 2025. Lobbying counts 30%: federal lobbying filings naming the fear in the past year. Public attention counts 30%, split evenly between news coverage and Wikipedia readership over the last 30 days. Each channel is log scaled to its leader, so a fear with a tenth of the leader's bills still registers rather than rounding to zero, and a fear with no Wikipedia article on file is scored on the channels it has. The raw counts behind every score are shown on its page and in the grid.

Bills, rules and orders

Every state and federal bill introduced since January 2025, plus federal rules and executive orders, that is substantially about artificial intelligence, algorithms, automated decisions, synthetic media or AI data centers. Bills are collected from Congress.gov and from Open States for all 50 states, DC and Puerto Rico. Federal rules and orders come from the Federal Register and are kept only when AI appears in the title or abstract.

Fears and controls

Each measure is read for the fears it cites and the controls it would impose, against fixed definitions of each. A first pass proposes labels. A second pass has to back every label with a quote copied from the measure's own title or summary, and that quote is then checked word for word in code. A label with no supporting quote of at least four words does not exist on this site. Every stored label is re-checked against the current rules on every run.

In their own words

Each quote is the line a label rests on, copied from the measure. The front page shows the five bills that hand out the most: ranked by the number of controls the bill carries, whether it names an office, whether it has passed, and how much of the controlling language sits in the quoted line itself. Companion bills with the same text count once. Bill titles written in capitals and lines under six words are left out.

Who's funding it

Federal lobbying filings whose issue descriptions name one of the tracked fears, over the past four quarters. Advocacy groups and foundations are ranked by dollars, because an organization whose stated purpose is a given risk is spending that money on the fear. Companies and trade groups are ranked by number of filings, because a trade group's filing reports one total for everything it lobbied on that quarter, so its dollar figure says how much it lobbies, not how much it spends on AI. Those dollar figures appear on each organization's page with that note attached.

Election money

Money raised and spent by political committees working on AI policy, counted by committee and kept out of the funding ranking, because a committee's spending does not say which side of a fear it is on. Individual donations appear on the receiving committee's page, each linked to its filing.

Who benefits

Government agencies and officials that a measure would hand new authority, jurisdiction or enforcement power over AI, counted by the number of measures. Legislatures and bare jurisdiction names are excluded.

Public attention

Daily Wikipedia readership of each fear's articles, and daily news volume from GDELT. Headlines come from publishers' own feeds, and a story counts for a fear only when every word that fear is defined by appears in it. The polling lines at the top of the front page are from the surveys they link to, are dated, and leave the page at eighteen months.