How many AI bills has each state introduced? A live count, updated hourly
A ranked, per-jurisdiction count of artificial intelligence bills, generated from a live corpus rather than written down once. Includes what the number counts, what it misses, and how to check it against the legislature yourself.
There is no official register of artificial intelligence legislation in the United States. No agency maintains one, no legislature tags its own bills that way, and the phrase does not appear in most of the bills that would belong on such a list. Every published count of state AI bills is therefore somebody’s inclusion rule, applied to somebody’s corpus, on some particular day.
Most of those counts are also stale. A number typed into a report in February is quoted for the rest of the year, through a spring session that may have doubled it.
The table below is generated when this page is served, from the same corpus the product runs on. It says what it counts, when it was taken, and which jurisdictions it covers.
| Rank | Jurisdiction | Bills matched | Share |
|---|---|---|---|
| 1 | New York | 96 | 9.0% |
| 2 | New Jersey | 90 | 8.5% |
| 3 | California | 67 | 6.3% |
| 4 | Illinois | 63 | 5.9% |
| 5 | Virginia | 41 | 3.9% |
| 6 | Hawaii | 36 | 3.4% |
| 7 | Massachusetts | 35 | 3.3% |
| 8 | Maryland | 35 | 3.3% |
| 9 | Oklahoma | 33 | 3.1% |
| 10 | Pennsylvania | 31 | 2.9% |
| 11 | Texas | 30 | 2.8% |
| 12 | Iowa | 29 | 2.7% |
| 13 | Florida | 26 | 2.4% |
| 14 | Missouri | 26 | 2.4% |
| 15 | Minnesota | 25 | 2.4% |
| 36 further jurisdictions | 399 | 37.6% |
New York leads with 96 matched bills, ahead of New Jersey (90) and California (67).
Reading the table without over-reading it
Three things are worth fixing in your mind before you quote any figure from it.
These are introductions, not laws. A bill that was filed, referred to committee and never heard again is in this count, and in most legislatures that describes the large majority of everything filed. Introduction volume measures legislative attention. It does not measure legal exposure, and the two diverge wildly: a state can file forty AI bills and enact none.
Volume is not the same as significance. One state’s forty bills may be forty variations on a disclosure requirement that never advances. Another state’s four may include a comprehensive act with a compliance deadline. Ranking by count answers “where is the attention” and not “where is the risk”, and a practice that treats the first as the second will spend its time in the wrong states.
The ranking moves with the session calendar, not with policy interest. Legislatures that convene in January and adjourn in spring will accumulate their entire year’s filings in a few months. A count taken in August and a count taken the following March are not comparable, which is why the table carries its own date rather than leaving you to guess.
Why counting these bills is harder than searching for two words
The reason there is no canonical number is that the subject resists a keyword.
Legislative drafters do not converge on vocabulary. A bill governing the same conduct might be drafted around “artificial intelligence”, “automated decision system”, “algorithmic discrimination”, “automated employment decision tool”, “generative artificial intelligence” or simply a defined term the act invents in section two and uses thereafter. A search for the obvious phrase returns a subset whose size depends on drafting fashion in that chamber.
The failure runs in both directions, and the second direction is the one that quietly corrupts a count. Plenty of bills mention artificial intelligence once, in a legislative findings section, before going on to regulate something else entirely. A keyword search cannot tell the difference between a bill that governs AI and a bill that gestures at it, so a naive count inflates.
Then there is the omnibus problem. A significant AI provision is frequently not in an AI bill at all. It is section 14 of an appropriations act, or an amendment to an existing consumer protection statute. Anything that counts whole bills by title is going to miss those.
The approach behind this table is to score each bill’s text against a written description of the topic, using embeddings, so that substance rather than vocabulary decides the match. That handles the drafting variation and the passing mention. It is a judgement rather than a fact, which is why the methodology is published and why the scope statements are described honestly: they are model-drafted and validated by measurement, not attorney-reviewed.
Per-jurisdiction corpus quality is not uniform, and it is published rather than smoothed over. Some legislatures make full bill text easy to retrieve and others do not, which affects how confidently a bill from that state can be matched on substance. The methodology page carries the per-state figures, generated from the same database as the table above.
Checking any row against the source
The count is only useful if you can get behind it, so every jurisdiction in the table links through to the bills themselves, and each bill links to the legislature’s own page. That page is the record. If a bill’s status here and its status on the legislature’s site ever disagree, the legislature is right.
For a specific question rather than a survey, going directly to the state is usually faster: most legislatures offer full text search over the current session, and Congress.gov covers federal bills well. What those sources will not do is let you ask the same question across fifty jurisdictions at once, which is the only reason a corpus like this one exists.
If you need this as a recurring answer
A count is a snapshot. The question underneath it, for most people who search for this, is a standing one: what changed in my practice area, in the states I care about, since I last looked.
That is what LawSignals is built for. Each practice area gets a Tracker over the same corpus this table is drawn from, and a monthly written issue that says what moved and why it mattered, in prose a client can be forwarded. The Trackers are free to browse, including the AI legislation Tracker behind this table.