Law of SIAI law & regulation tracker
Language: English

Methodology

AI-generated. This page was written and translated automatically by AI and checked by software against the cited sources. No human editor reviewed it. Not legal advice.

Law of SI is produced by an automated pipeline. This page explains each step and what the checks can and cannot catch.

1. Sources

We monitor official, primary sources through their public feeds and interfaces, for example: the US Federal Register, US federal agencies’ press releases, state legislatures, the EU’s EUR-Lex and the European Commission, data-protection authorities, the UK government and Parliament, and official government sites in China, Japan, South Korea, Brazil and elsewhere. Court cases are tracked through public court records, mainly the CourtListener database of US federal court dockets, and official court sites elsewhere.

A news item is published only if it has a primary source. News reports may be listed as additional coverage, but they never replace the primary source.

Our crawler identifies itself as lawofsi-bot and respects robots.txt and rate limits. See /bot.

2. Frequency

Sources are checked about every 30 minutes. A new item normally appears within a few hours of the underlying document being published. Laws, bills and court cases already in the trackers are re-checked against their official pages every day. The weekly digest is generated every Monday.

3. Writing

An AI model (Anthropic’s Claude) reads the primary source and fills in a fixed set of fields: date, jurisdiction, type of development, organisations involved, and a short neutral title, summary and “why it matters” note. It is instructed to use only what the source says.

Some fields are never written by a model:

4. Checks before publishing

Each item must pass all of these checks; if any fails, the item is held back and retried later, and dropped after 3 failed attempts:

  1. Rule checks — required fields, valid dates, identifiers present in the source, the source on a list of official domains.
  2. Independent cross-check — a model from a different provider (OpenAI) compares the summary with the source and flags anything not supported by it, any statement attributed to the wrong party, and any allegation written as if it were proven.
  3. Duplicate and consistency checks — reports of the same development are merged; conflicting sources lead to more cautious wording or a delay.
  4. Build checks — the site is not published if pages are empty, untranslated text remains, or the number of items drops suddenly.

In court cases, what a party claims is reported as an allegation (“alleges”, “claims”) unless a court has decided the point. Names of private individuals are not shown.

5. Translation

English is the source language. Translations into Chinese, Spanish, French, German, Japanese and Italian are made by an AI model using a legal glossary, then checked by software (for example, that numbers, identifiers and dates match the English text and no paragraphs are missing). A translation that fails is not published until it passes.

6. Limits

Automated checks reduce errors but cannot remove them. Summaries are short and leave out detail; status information can lag behind the official record; and machine translation of legal terms is imperfect. Always rely on the linked primary source, and see the disclaimer.

7. Labelling

Every page carries a visible “AI-generated” label right below its title, in each language. Feeds and downloadable data mark every record as AI-generated and not human-reviewed, and pages are served with a machine-readable AI-Disclosure header. This follows the labelling practice set out for AI-generated text in the EU AI Act (Article 50) and in China’s rules on labelling AI-generated content.

8. Corrections

Anyone can report an error from any page. Reports are checked automatically against the primary source; see the corrections policy.

Updated: