How the numbers work

Every number on this site is counted from public records, and every bill, filing and committee behind it links to its source. The data and the code are public, and the site updates itself through the day.

The 17 fears

The report follows 17 fears: loss of control, the China race, job loss, power bills and data centers, kids and chatbots, AI therapists, deepfakes, algorithmic bias, AI deciding care, privacy and personal data, AI taking creators' work, AI surveillance, algorithmic pricing, AI companions, AI scams, AI misinformation and self-driving vehicles. They are the 17 that scored highest on the Fear Index out of 25 fears measured on 25 September 2026, drawn from national polls on what worries people about AI, the subjects that recur in AI bills, and the fears named in lobbying filings and news coverage. Each fear has one definition, used for bills, lobbying, news and Wikipedia alike, and the list holds still so that a score means the same thing from one day to the next.

The fears measured and left off are AI cyberattacks, bioweapons, AI in schools, AI weapons, addictive algorithms, AI monopolies, the AI bubble and AI slop. They are scored every day the same way, and the weekly review names any that comes to outscore a fear on the list.

The Fear Index

Each fear gets a score from 0 to 100. Legislation counts for 40%: the bills, resolutions, rules and orders since January 2025 that cite the fear. Lobbying counts for 30%: the federal lobbying filings that name it in the past year. Public attention counts for the last 30%, split evenly between news coverage and Wikipedia readership over the past 30 days. On each channel a fear is measured against the leading fear on a log scale, so one with a tenth of the leader's bills still registers. Each fear's page shows the counts behind its score.

What counts as a measure

State and federal bills and resolutions introduced since January 2025, or filed for a session that began then, plus federal rules and executive orders. A measure counts when it is about AI, algorithms and automated decisions, deepfakes and other synthetic media, data centers, or self-driving vehicles. Data center bills count whether or not they mention AI, because AI is driving the data center boom, and so do bills on self-driving vehicles, because an AI is the driver. A bill whose title names AI, data centers or self-driving vehicles always counts, and one that mentions AI only in passing, such as an agriculture bill that lists it among other technologies, does not.

Bills come from Open States and Congress.gov, which together cover Congress, all 50 states, DC and Puerto Rico. Rules and executive orders come from the Federal Register. The record is checked against a list of AI laws known from other sources, and a missing one is added. A bill carried over into a later session counts once. Every bill links to its legislature's own site.

State by state

The map shades each state by its AI measures that would put AI under new government control, on a log scale against the state with the most. It can switch to what is already law or to a single fear.

Passed and not passed

Where a measure stands comes from the last action its legislature recorded. Passed means enacted: signed, chaptered, or law without the governor's signature. Sent to the governor, or to the president for a federal bill, means the legislature passed it and nobody has signed or vetoed it since. Did not pass, vetoed and withdrawn mean the record shows it stopped. Everything else is not passed. A federal rule is proposed or final, and an executive order is signed. A resolution that passes reads adopted and is not counted as law.

Fears and controls

Each measure is labeled with the fears it cites and the government controls it would create, each under a fixed definition. Every label rests on a quote copied word for word from the measure's official title or summary and is checked against that text. Many legislatures publish no summary, or a single line, so a law whose summary runs under 25 words is also read from its enacted text, and its labels can quote that. A bill still moving with no summary is labeled from its title alone and can carry fewer labels than its text would support. A label that doesn't hold up is removed, and one that does stays when the measure is read again. Only what the measure itself would do counts, not its description of laws already on the books.

The controls are licenses required to build AI, caps on who can train models, limits on sharing open models, ID and age checks for users, mandatory reporting to government, labels and notices of AI use, bans and limits on uses of AI, new powers for government agencies, export controls on models and chips, state laws overridden by Washington, and moratoriums and limits on data centers. A measure counts whether or not it cites a fear, and so does every control it would create.

Rules for the government's own use of AI, such as an inventory of the systems state agencies use, standards for what they buy, or policies for AI in schools and policing, are not controls. They bind the government, not the people who build, sell or use AI. A measure whose only rules are of that kind counts as an AI measure but not as one that adds government control.

In their own words

The quotes on the front page are copied from each bill's official summary, or from a law's text where the summary is too short to quote, the lines its labels rest on. They come from the bills that would hand out the most power, ranked by how many controls they carry, whether they name an office, and whether they have passed. Companion bills with the same text count once.

Who's funding it

Federal lobbying filings from the last four reported quarters whose issue descriptions name one of the fears. Advocacy groups and foundations are ranked by dollars. Companies and trade groups are ranked by number of filings, because each of their filings reports one total for everything they lobbied on that quarter. A client that lobbies in-house already reports what it paid outside firms, so that money counts once.

Election money

Money raised by political committees working on AI policy, found by name and by searching the FEC's register for committees with AI, artificial intelligence, machine learning, deepfake, algorithm or data center in their names. It is kept apart from the lobbying ranking, because a committee's money doesn't show which side of a fear it is on. Donors of $1,000 or more are named on each committee's page, linked to their filings.

Who benefits

The agencies and officials that measures carrying a control would give new power over AI, data centers, or the people and companies who build or use AI, ranked by how many of those measures name them. Legislatures don't count, and neither does an office that would only oversee the government's own use of AI.

Public attention

News volume from GDELT and Wikipedia readership of each fear's articles, over the past 30 days. The headlines come from the publishers listed below, and one counts toward a fear only if it contains every term that defines the fear. The quote at the top of the front page is the headline about the leading fear most like the rest of that week's coverage. The polls on the front page link to their surveys and come down after 18 months.