US State AI Law API: statutes, effective dates and official citations

The free trackers are wrong, and this API ships the receipts.

Every record here was verified against the state's own legislature or codified statute, not against another tracker. Where the trackers disagree with the official record, the disagreement is in the dataset as a queryable finding, with the source that settles it. There are 37 of them in the current corpus.

Here are three, quoted from the data you get back:

Colorado AI Act reported as in force: it was repealed before it ever operated. As of 2026-08, widely-syndicated summaries (including the Wikipedia 'Colorado AI Act' article and multiple vendor compliance blogs) still state that the Colorado AI Act 'commenced on June 30, 2026' and describe its duty of care, risk-management programs and impact assessments as live obligations. Verified against the official Colorado General Assembly record: SB 24-205's duties were delayed to 2026-06-30 by SB 25B-004 (signed 2025-08-28), AG enforcement was stayed by joint motion in xAI v. Weiser, No. 1:26-cv-01515 (D. Colo.) on 2026-04-27, and SB 26-189 (Chapter 131, signed 2026-05-14) repealed and reenacted the entire article with a disclosure-based framework whose compliance date is 2027-01-01. A buyer relying on the incumbent summaries would build to a risk-management and impact-assessment regime that no longer exists in Colorado law.

New York RAISE Act: the signed text is not the operative text. The official nysenate.gov page for S6953B (Chapter 699 of 2025, signed 2025-12-19) still states the effective date as 'ninety days after it becomes law'. The operative statute is the chapter-amended version: S8828, signed 2026-03-27 as Chapter 96, which moved the effective date to 2027-01-01, replaced the compute-cost coverage test with a $500 million revenue test, reduced penalties from $10M/$30M to $1M/$3M, and created an oversight office inside the Department of Financial Services. Any tracker that captured the law at signing is wrong on the date, the covered population, the penalty and the regulator simultaneously.

Connecticut tracked as 'SB 2': the enacted law is SB 5 / Public Act 26-15. Connecticut SB 2 is the bill number carried by every 50-state AI survey from 2024 through early 2026, because SB 2 passed the Connecticut Senate in both 2024 and 2025. It never became law in either year. The comprehensive act that WAS enacted is SB 5 of the 2026 session, signed 2026-06-02 as Public Act 26-15 (the CART Act), verified against the Governor's official press release. Trackers keyed to bill numbers rather than session laws will either show Connecticut as having no comprehensive AI law or show a dead bill.

Note what those three have in common: a careful researcher reading the official page still gets New York wrong, because the official page itself displays the superseded text. That is the work this API is doing for you.

Why this exists

The closest incumbent, AILawsByState, sells a $39/month human-readable tracker and puts API access behind a $499/month Enterprise tier. Everyone else in this space sells a PDF chart, a law-firm newsletter, or a seat licence.

Nobody sells the developer API. That slot is empty, and this fills it: JSON in, JSON out, stable ids, official citations, one call.

What you get back

Enacted laws and adopted regulations	57
Jurisdictions covered	24 (23 states + New York City)
Categories	7
Staged effective dates, individually queryable	122
Recorded tracker discrepancies	37
Records whose official URL was actually fetched and content-matched	116 of 118
Pending bills tracked	3

Every law record carries: state, citation of record, codified cite, popular name, category, plain-English summary, status, every staged date with what happens on it, who it covers, the obligations as a list, enforcement and penalties, amendment history, the official URL, whether that URL was verified, and a review flag marking legally complex records, with the reason in notes.

Categories: comprehensive-ai, ncii-digital-replica, health-ai, deepfakes-elections, ai-hiring, chatbot-disclosure, gov-procurement.

Modes

Mode	What it does	Billing
search_laws	Filter by state, category, status, in-force, date window, who-it-covers, or full text across summaries and obligations	1 event per call
effective_calendar	Give it a date window, get back the obligations that start (or end) in it	1 event per call
law_detail	One law in full, plus the tracker discrepancies that bear on it	1 event per call
discrepancies	Where the public trackers are wrong, and what settles it	1 event per call
coverage	Counts, freshness, corpus version, and the honest coverage map	free

One event per call, not per row, a calendar query that returns forty obligations costs the same as one that returns none.

The compliance calendar, in one call

{"mode": "effective_calendar", "from_date": "2026-08-01", "to_date": "2027-01-31"}

Returns one entry per staged date, the date, what happens on it, and the law's obligations, because a single act stages several. Colorado's replacement act stages four dates; California's CPPA ADMT regulations stage seven, out to 2030. "What lands on me in Q1" is an indexed query here, not a pile of statutes to read.

Stable ids, because everything else moves

law_id is keyed on (state, citation of record). It survives what actually changes: status, effective dates, penalties, the regulator, even the codified cite. Measured on this corpus, Washington's citation moved from RCW ch. 42.62 to ch. 29B.35 on 2026-01-01; Utah's disclosure duty moved from § 13-2-12 to § 13-75-103 on 2025-05-07 and the old section was expressly repealed. Store the law_id, not the cite.

Repealed laws stay in the dataset, on purpose

status is one of effective, upcoming-effective-date, amended, enacted, enjoined_in_part, repealed, and each carries a coarse in_force flag (yes / partial / not_yet / no). Filter them out when you want today's duties, but "this is no longer the law" is an answer worth having, and it is the answer the free trackers are getting wrong about Colorado right now.

enjoined_in_part is flagged rather than resolved, because it cannot be resolved into a boolean: California's election-deepfake package has four provisions with three different fates and an undecided Ninth Circuit appeal.

Coverage, stated honestly, and why we never say "no law"

Call coverage (free) for the live map. It reports every jurisdiction as either covered or unknown, and there is deliberately no third value.

We do not claim that any state has no AI law. 41 jurisdictions are covered; the other 28 are unswept, not empty. That distinction is the whole product: a compliance buyer who reads "no law in Ohio" and ships accordingly has been handed a false negative, and Ohio has two AI bills in this dataset's own pending table. Where we already hold evidence that an empty reading would be wrong (Arizona, Ohio, Georgia, New Jersey, Massachusetts, Virginia and more), the unknown entry says so and names the lead.

The same rule holds inside a covered state: a category we did not sweep there comes back unknown, not empty. Per-category completeness is published in the same call, in the category's own terms, for example, gov-procurement is 5 states, not 50, and says so; deepfakes-elections is explicitly non-exhaustive (Kentucky SB 4 contains a synthetic-media election provision and is not yet recorded as an election record, a gap we found in our own data and published rather than hid).

Freshness is auditable too: every response carries corpus_version, corpus_as_of and the corpus file's sha256.

Refresh: status churn is what goes stale, not new enactments

Five laws in this corpus produced 15 status changes in about two years. Colorado SB 24-205 alone changed four times: enacted, delayed, judicially stayed, repealed and reenacted. A quarterly refresh would have served the wrong Colorado answer for up to three months, at the exact moment the market cared most. Records with a future staged date, an upcoming-effective-date status or an enjoined_in_part status are the ones that move, and they are queryable as such.

Typical uses

• A compliance calendar inside your own product, keyed to the states you sell in
• Gating a feature by jurisdiction (an AI therapy feature is lawful in Utah with disclosures and prohibited outright in Illinois and Nevada, the category name "chatbot disclosure" hides that, and the obligations text does not)
• Due diligence and policy research with citations you can click through
• Feeding an LLM structured, cited law instead of letting it recall AI statutes from memory

Honest limits

• Not legal advice. Structured legal information with official citations. Verify against the linked official source before acting; 84 of 118 records carry a review flag for legal complexity, and each says why in its notes.
• 41 jurisdictions, not 51. Everything else returns unknown, never "no law". See the coverage section above. this is the limit we are most insistent about, because it is the one that could hurt you.
• Two records could not have their official URL fetched (Utah SB 149, le.utah.gov 404s the 2024 enrolled-bill path; Indiana HEA 1133, iga.in.gov is a JavaScript shell). Both say so in their own url_verified field and notes, and both were established from other official texts.
• Connecticut is the weakest verification in the corpus: cga.ct.gov serves an incomplete TLS certificate chain and could not be fetched at all, so both Connecticut records rest on the Governor's official press release plus secondary law-firm analysis. Published as a caveat, not buried.
• Some codified cites are article-level rather than section-level, and each such record says so.

Methodology: every record is built from, and checked against, the official primary source, and each source's URL and as-of date are published so you can verify any result yourself. Public trackers (NCSL, Public Citizen, IAPP) were used as leads only; every citation of record is an official source.

Support: friendlyapidev@gmail.com

---

Terms of Use

Effective 9 August 2026. These Terms govern this and every other data service published by the Provider. "The Provider" means the operator of the Service, reachable at friendlyapidev@gmail.com. "The user" means anyone who accesses the Service, paid or free. By using it you agree to them.

§1's consumer-reporting provisions and §5's prohibition on decisions about individuals are directed principally at services that return records about named people.

1. What this service is

The Service provides structured data compiled from public government sources. It is an information product. It is not legal advice, not a compliance determination, and not a background-check or consumer report.

Nothing in the Service constitutes a verdict about any person or entity. The Service returns candidate matches with confidence scores and source provenance, which a competent human must review before any action is taken.

The Provider is not a consumer reporting agency, and nothing the Service returns is a "consumer report" or "investigative consumer report" within the meaning of the Fair Credit Reporting Act (15 U.S.C. § 1681 et seq.) or any state analogue. The Service is not assembled or evaluated for the purpose of establishing a consumer's eligibility for employment, credit, insurance, or housing, and the Provider does not follow the procedures the FCRA requires of consumer reporting agencies. Permitted uses are set out in §5.

2. No warranty

THE SERVICE AND ALL DATA ARE PROVIDED "AS IS" AND "AS AVAILABLE", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY WARRANTY OF ACCURACY, COMPLETENESS, CURRENCY, MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT.

Specifically, and without limiting the above, the Provider does not warrant that:

• the data is complete, current, or free from error;
• every relevant record has been collected, or that any particular jurisdiction, list, or record is included;
• a "no match" result means a person or entity is not subject to any exclusion, sanction, action, or legal obligation;
• source publications will remain available, unchanged in format, or on any particular schedule.

3. The data comes from third parties we do not control

All data is derived from public government publications. Those publishers control what they publish, when, in what format, and whether it is accurate. They change formats without notice, take sites offline, and publish records containing their own errors. The Provider reproduces and normalises what those sources publish and does not independently verify the truth of any underlying record.

Each result identifies the source it came from, and the Service publishes the as-of date and official URL of every source it carries, so any result can be checked against the government publication it derives from.

4. The user's verification duty

The user is solely responsible for verifying any result against the official primary source before relying on it or acting on it, and for obtaining professional advice appropriate to their circumstances.

The Service's published coverage statement is part of these Terms. It states which jurisdictions are collected automatically, which are published by their source in a form the Service does not collect, and which publish nothing at all. A user who acts on the Service without reading it does so at their own risk.

5. Acceptable use

Permitted. Query the Service, integrate it into your own product, and display individual results to your own users, including as part of a compliance workflow you operate.

Prohibited. The user shall not:

1. Use the Service, in whole or in part, as a factor in any decision about an individual's employment, engagement, retention, promotion, credit, insurance, housing, or eligibility for any benefit. The Service is not a consumer report (§1). Screening obligations that arise from law or regulation must be discharged against the official primary source, and any decision about a person must rest on that verified source and the user's own judgement, not on this Service.
2. Extract, copy, or redistribute the dataset in bulk, or use it to build, train, or populate a competing dataset or data feed. (Querying the Service and showing results to your own users is expressly permitted above.)
3. Use the Service in violation of any law, or to re-identify, harass, or defame any individual appearing in the data.

A record in these datasets means only that a government body published it. It is not a finding by the Provider about any person.

6. Limitation of liability

TO THE MAXIMUM EXTENT PERMITTED BY LAW, THE PROVIDER SHALL NOT BE LIABLE FOR ANY INDIRECT, INCIDENTAL, SPECIAL, CONSEQUENTIAL, EXEMPLARY, OR PUNITIVE DAMAGES, OR FOR ANY LOSS OF PROFITS, REVENUE, DATA, GOODWILL, OR BUSINESS OPPORTUNITY, ARISING OUT OF OR RELATING TO THE SERVICE, WHETHER IN CONTRACT, TORT (INCLUDING NEGLIGENCE), OR ANY OTHER THEORY, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.

THE PROVIDER'S TOTAL AGGREGATE LIABILITY ARISING OUT OF OR RELATING TO THE SERVICE, IN THE AGGREGATE AND FOR ALL CLAIMS COMBINED, SHALL NOT EXCEED ONE HUNDRED US DOLLARS (US$100).

The user acknowledges that this limitation is a material part of the bargain, that the fees charged reflect it, and that the Provider would not offer the Service on these terms without it.

Two notes on why the cap is a flat figure rather than a multiple of fees. It is never zero. A cap of zero for a user who paid nothing invites the argument that the clause is illusory, and a court that agrees may strike the whole limitation rather than read it down. And it is fixed rather than computed, so there is nothing to reconstruct from billing records years later.

Nothing in these Terms excludes or limits liability that cannot lawfully be excluded or limited. If any part of this section is held unenforceable, the remainder continues to apply to the fullest extent permitted.

7. Indemnity

The user shall indemnify and hold the Provider harmless from any claim arising from the user's use of the Service, including any decision made in reliance on it and any breach of §5.

8. Availability and changes

The Service may change, add, or remove data sources and coverage at any time, and may be unavailable. Changes to coverage are reflected in the published coverage statement.

9. Governing law

These Terms are governed by the laws of the State of New York, without regard to conflict-of-laws principles. The parties submit to the exclusive jurisdiction of the state and federal courts located in the State of New York, and each waives any objection to that venue.

Disputes are resolved in court; these Terms do not require arbitration.

Pricing

Pay per event, no subscription. Prices fall on higher Apify plans.

Event	What triggers it	Price
law_search	One search across the corpus	$0.10 (from $0.06)
calendar_query	One compliance-calendar window	$0.25 (from $0.15)
law_detail	One full law record	$0.05 (from $0.03)

Checking the whole US map once a week costs a few dollars a month.

10. Contact

friendlyapidev@gmail.com

Support is handled by email. There is no telephone support and no guaranteed response time.

---