Methodology
How we verify
AI Law Radar tracks 33 obligations across 13 jurisdictions, with 60 jurisdictions shaded on the world map. Every fact is built to one standard: it links to a primary source, carries the date it was last checked, and declares how confident we are.
Primary sources only
Every row links to a primary source — an official journal, a statute or bill portal, or a regulator's own notice — never a secondary summary or a news write-up. Where an instrument exists only as a bill, we link the legislature's tracking page so you can see its exact stage for yourself.
Checked dates & confidence flags
Each obligation carries a checked date: the day a human last reconciled the row against its source. It also carries a confidence flag — high, medium or low. Anything in trilogue, pre-enactment, or pending official-journal publication is marked low and dated to its last legislative motion, so you never mistake a proposal for binding law.
Daily re-verification
An automated research pass re-checks the dataset on a schedule and stages any changes; a deterministic validator (standard-library only, no model in the loop at the gate) blocks a deploy unless every structural and source-integrity check passes. When a change is confirmed against its primary source, it lands in What changed and the site redeploys itself. The "last verified" date in the masthead is derived from the newest checked date in the data — not a wall-clock timestamp.
The five-status scale
On the map and the directory, each jurisdiction is shaded by its strongest in-force instrument:
- Comprehensive Comprehensive law
- Binding Binding sectoral
- Proposed Proposed / drafting
- Guidance Guidance only
- None No AI-specific law
Scope
33 obligations across 13 deeply-tracked regions — the EU, the US (federal and the CA, CO, IL, NY, NYC and TX state rules), the UK, China, South Korea, Japan, Vietnam, Peru, India, Canada, Mexico, Brazil and the UAE — plus 60 jurisdictions shaded on the world map at status level. We cover horizontal AI acts, binding sectoral rules (deepfake labelling, automated-decision transparency, frontier-model safety) and notable proposals.
Limitations
This is a tracker, not a compliance product. We record the existence, scope and timing of obligations; we do not assess whether a specific system is in scope, nor do we capture every delegated act, guideline or local enforcement practice. Thresholds, exemptions and definitions in the underlying instruments are summarised, not reproduced. Dates can move — especially for proposals and items pending publication — which is exactly why every row is dated and flagged.
Not legal advice. AI Law Radar is an information service. Nothing here is legal advice or a substitute for qualified counsel in the relevant jurisdiction. Always verify against the linked primary source before you act.