Data
Corpus scope and coverage
What TwinEthos encodes, where it applies and how it is checked, at a glance. Every figure is computed from the corpus each time the site is built: data as of 1 Nov 2026.
- 220 of 221In-scope enacted AI lawsencoded so far
- 353Binding AI-law rulesin 55 jurisdictions: 17 outside the US, 38 in the US
- 48Related guardrailsTwinEthos recommendations, not law
- 180 of 180Related controlsstate the guard to add
- 553Official source documents
A law is usually captured from several documents: its codified sections, the enacted bill, implementing regulations, legislative records and gazette notices. The sources also include standards, guidance and laws outside the in-scope count.
371 statute texts · 71 regulations · 30 legislative records · 27 official gazette issues · 23 guidance documents · 12 agency publications · 1 court decision · 1 translation · 17 not yet classified
“In scope” is a TwinEthos judgment of whether a guard in application code can address a legal rule; it is never a legal conclusion. The laws include AI-adjacent privacy and biometric laws. No rule has been reviewed by a lawyer: use it as information only, not legal advice.
At a glance
Laws by jurisdiction
In-scope enacted laws, the 20 jurisdictions with the most. Select a bar to open it in Explore.
Also: Nevada (3), New Hampshire (3), Washington (3), Alabama (2), Arizona (2), Brazil (2), Delaware (2), Dubai International Financial Centre (2), Ecuador (2), Florida (2), France (2), Georgia (2), Hawaii (2), India (2), Iowa (2), Italy (2), Kazakhstan (2), Kyrgyzstan (2), Michigan (2), New Jersey (2), New York City (2), Nigeria (2), Peru (2), Turkey (2), United Arab Emirates (2), United Kingdom (2), Vermont (2), Arkansas (1), Chile (1), El Salvador (1), Germany (1), Ghana (1), Greece (1), Idaho (1), Indonesia (1), Japan (1), Kentucky (1), Kenya (1), Louisiana (1), Mauritius (1), Mexico (1), Minnesota (1), New Mexico (1), North Dakota (1), Oklahoma (1), Ontario, Canada (1), Philippines (1), Portland, Oregon (1), Quebec, Canada (1), Russia (1), Saudi Arabia (1), Serbia (1), Singapore (1), South Africa (1), South Carolina (1), South Dakota (1), Switzerland (1), Uganda (1), Uruguay (1), Uzbekistan (1), Vietnam (1), Virginia (1), Wisconsin (1) (93 laws in 63 more jurisdictions).
Rules by lane
The core AI group. AI-adjacent privacy and biometric law adds 147 rules, counted apart. Select a lane to open its rules in Explore.
Coming into force
Dated obligation starts in binding core AI law, by quarter: 149 starts from 129 rules (Coming into force in the explorer adds AI-adjacent law). Select a quarter to open its dates.
Totals by jurisdiction
Laws in scope and encoded come from the coverage registers; rules are the encoded provisions that bind in each place, by lane. A rule that binds several places counts in each, and once in a total. A city's laws and rules are listed apart from its state's.
Countries and regions outside the US
40 jurisdictions: 73 of 74 in-scope laws encoded; 110 binding AI-law rules (88 in force, 22 not yet in force or stayed), 19 standard and soft-law rules and 88 AI-adjacent rules.
US states and cities
42 jurisdictions: 136 of 136 in-scope laws encoded; 237 binding AI-law rules (141 in force, 96 not yet in force or stayed) and 46 AI-adjacent rules.
US federal
1 jurisdiction: 11 of 11 in-scope laws encoded; 6 binding AI-law rules (6 in force, 0 not yet in force or stayed), 8 standard and soft-law rules and 13 AI-adjacent rules.
| Jurisdiction | Laws in scope | Encoded | Binding AI-law rules | In force | Not yet in force or stayed | Standards and soft law | AI-adjacent rules |
|---|---|---|---|---|---|---|---|
| United States, federal (US) | 11 | 11 | 6 | 6 | 0 | 8 | 13 |
Sliced views
By AI feature
A rule tagged with several features counts under each.
| AI feature | Binding AI law | AI-adjacent law | Standards and soft law | Recommended guardrails |
|---|---|---|---|---|
| Chat or assistant | 145 | 31 | 41 | 30 |
| Answers from your documents (RAG) | 7 | 10 | 22 | 18 |
| Agents that use tools or take actions | 27 | 11 | 49 | 31 |
| Decisions about people (hiring, credit, insurance, health) | 118 | 77 | 38 | 11 |
| Generated text, images, audio or video | 110 | 39 | 22 | 16 |
| Classification, scoring or biometrics | 98 | 82 | 13 | 9 |
| Recommendations and feeds | 5 | 18 | 2 | 4 |
| Embeddings and vector search | 5 | 19 | 2 | 3 |
| Any AI feature | 55 | 19 | 11 | 3 |
By sector
Rules whose domain goes beyond AI itself.
| Sector | Binding AI law | AI-adjacent law | Standards and soft law | Recommended guardrails |
|---|---|---|---|---|
| Data | 38 | 109 | 6 | 0 |
| Health | 77 | 14 | 0 | 0 |
| Cyber security | 1 | 1 | 43 | 0 |
| Insurance | 31 | 2 | 0 | 0 |
| Employment | 19 | 2 | 0 | 0 |
| Financial services | 0 | 15 | 0 | 0 |
| Ethics | 1 | 0 | 6 | 0 |
By lane
| Lane | Rules |
|---|---|
| Binding law — in force | 235 |
| Binding law — not yet in force or stayed | 118 |
| Standard / soft law | 89 |
| TwinEthos recommendation (not law) | 48 |
| Core AI group | 490 |
| AI-adjacent law: in force | 125 |
| AI-adjacent law: not yet in force or stayed | 22 |
| AI-adjacent group, counted apart | 147 |
By year in force or coming into force
Binding AI-law rules by the year their obligations first apply.
| Year obligations first apply | Binding AI-law rules | In force | Not yet in force or stayed |
|---|---|---|---|
| 2019 | 1 | 1 | 0 |
| 2020 | 5 | 3 | 2 |
| 2021 | 1 | 1 | 0 |
| 2022 | 4 | 3 | 1 |
| 2023 | 8 | 8 | 0 |
| 2024 | 20 | 20 | 0 |
| 2025 | 76 | 75 | 1 |
| 2026 | 138 | 124 | 14 |
| 2027 | 96 | 0 | 96 |
| 2028 | 3 | 0 | 3 |
| 2029 | 1 | 0 | 1 |
How the data is checked
Every build checks each rule, guardrail, control and source against TwinEthos's own quality bar (provenance, verbatim text, dates, applicability, detectors and the guard to add); it is not a legal review.
Corpus as of 2026-11-01: 99.9% of 10,998 applicable checks pass, and 1,380 of 1,389 artifacts meet every criterion.
Global metrics
From the corpus audit of 2026-11-01, against TwinEthos's own audit standard: a quality bar for provenance, dates, applicability, detectors and guards, not a legal review. No rule has been reviewed by a lawyer yet. How the data is checked. What is encoded and what is not.
Corpus audit99.9%of 10,998 checks pass
against TwinEthos's audit standard; 1,380 of 1,389 rules, guardrails, controls and sources meet every criterion.
Verbatim provisions2,850stored provisions found word for word in the official text
1,743 in the live document by the weekly watcher, 1,107 in the captured copy; 19 with no check recorded yet. None was checked and not found.
Archived sources553official-source captures kept
Each matched to the hash recorded when it was taken (for a paywalled standard, the receipt rather than its text).
Controls with a guard180 of 180controls state the guard to add
What it does, where it goes in the code and what a reviewer checks.
Global coverage353binding AI-law rules across 55 jurisdictions worldwide
17 international (outside the US) and 38 in the US (federal, state and city), plus 48 recommended guardrails (TwinEthos opinion, not law) backed by 51 graded real incidents (real-world AI incidents on the public record, each graded by the best source behind it: a primary record, the press of record or the trade press).
| Artifact | Artifacts | Meet every criterion | Checks passed | Score |
|---|---|---|---|---|
| Binding-law rules | 500 | 497 | 4,977 of 4,980 | 99.9% |
| Standard and soft-law rules | 89 | 85 | 879 of 883 | 99.5% |
| Recommended guardrails | 48 | 48 | 566 of 566 | 100.0% |
| Controls and control families | 198 | 198 | 719 of 719 | 100.0% |
| Sources | 554 | 552 | 3,846 of 3,850 | 99.9% |
Stored provisions found word for word in the official text: 2,850 (1,743 in the live official document by the weekly watcher, 1,107 in the captured document by a local check); 0 checked and not found; 19 with no check recorded yet. A check with no evidence in the repository never counts as a pass (5 checks today), and no result says that a rule states the law correctly or that any code meets any law.