Methodology · data as of 2026-10-21

How the observatory works.

What this is

A map of how intellectual property law is responding to artificial intelligence: court and office decisions, legislation, and proposals from parliaments, governments, universities and professional bodies, in one place. It is descriptive. Havenmark takes no position on what the law should be; where the site ranks or summarises, it reports breadth of support, not merit.

Sources

94 curated decisions and acts. This is not a full case database. We select the decisions that set or shift precedent: rulings of apex and appellate courts, the first decision on a question in a jurisdiction, decisions that break with the trend, and the office decisions and legislation practitioners rely on. Each links to the official text where one is available, and to a reputable summary.

59 policy papers. Reports, resolutions, bills and consultations from policymakers; peer-reviewed articles, books and studies by recognised scholars; and resolutions of professional bodies. We link only to official or openly available copies.

3,638 EU judgments.Every judgment of the Court of Justice and the General Court since 2010 tagged with an intellectual property subject in the EU Publications Office’s official data (CELLAR). Labels come from each judgment’s official keywords, not from our reading. EUR-Lex data is reused under Commission Decision 2011/833/EU.

How positions are coded

Each decision or paper is coded against one question, on a five-point scale between the question’s two poles. Only the core holding or main recommendation counts, not side remarks; a source that discusses a question without taking a position is recorded but not placed. A judgment with separate holdings on two questions (GEMA v. OpenAI on the model and on outputs) is coded on each.

Descriptive work, such as comparative surveys and mapping studies, is shown as evidence, not as a position: it is listed on its question but not placed on the scale, and not counted as backing a proposal.

For AI training, for example: −2 permission is needed; −1 training is allowed only with payment, or licensing is the main route; 0 allowed with limits (an opt-out, case by case); +1 generally allowed, with limits only at the edges; +2 freely allowed. The other questions follow the same pattern between their own poles. Where a source’s lean within a middle position is not clear from its text, it stays at 0.

Every record has its own page with a stable link, the position on each question it answers and why it sits there, and a log of later changes to the record. Where the reasons for a position between the poles are not yet written up, the page says so.

Where the link between a decision and a principle is arguable, it is marked with a question mark and not counted.

Authority and good law

Decisions carry their level (legislation, apex, appellate, first instance, office, guidance) and their status: Current, Ongoing (under appeal), Narrowed or Overruled, with the later decision that changed it. Narrowed and overruled decisions are left out of every consensus and direction measure. Statuses are checked by hand, not by an automated citator; see the date below.

Where courts agree

A principle counts as agreed when good-law decisions in at least two regions support it and none in our data contradicts it. Strength is shown by level: apex courts and legislation, appellate courts, then lower courts and offices.

Law and proposals are kept apart

The observatory separates what the law is from what people propose it should be. Court decisions, legislation and office practice form the law layer; papers from policymakers, academia and professional bodies form the proposals layer. In the overview the law sits above each line and proposals below it, and each layer has its own verdict.

The law is read jurisdiction by jurisdiction. In each jurisdiction only the most authoritative good-law source that takes a position counts: legislation and apex courts, then appellate courts, then first instance, then office practice, newest first on ties. Ten lower-court rulings in one country therefore do not outweigh one apex ruling elsewhere. Law marks are drawn larger for higher authority, and as rings for office decisions and guidance. A refusal to hear an appeal, such as a denial of certiorari, is recorded as a development but not as a position, since it decides nothing on the merits.

Each layer gets a verdict: Too few under three sources (no claim either way); Settled when at least three quarters sit on one side or in the middle; Split when at least a quarter sit on each side; otherwise Leaning. Within each question, every source is still shown as its own mark; nothing is averaged.

Ranking proposals

Proposals are grouped into shared themes. Across all sources, each theme is scored by the share of each group’s institutions that back it, averaged so that policymakers, academia and professional bodies weigh the same; a large pool in one group cannot outvote the others. Two papers from the same institution count once. Ties go to the theme backed in more regions.

Limits

The selection is curated, not exhaustive, and leans toward sources available in English. Summaries are simplified and may be drafted with AI assistance before editorial review; always verify against the primary source. Nothing here is legal advice.

Codebook

The open dataset (CSV, JSON) has one row per decision, act or paper, with these fields:

id
Stable identifier; the record's page is /pro/decisions/{id}
title
Short name of the decision, act or paper
kind
court, office, act or policy (non-binding proposals, studies and reports)
body
Court, office, legislature or authors
jurisdiction
Jurisdiction it belongs to
region
Region used for grouping
date
ISO date, or year-month when the day is uncertain
citation
Citation to look it up by
issue
The question it is coded on
answer
yes, partly, no, or none (takes no position)
position
Place on the question's five-point scale, from -1 (left pole) to 1 (right pole); empty when it takes no position
level
legislation, apex, appellate, first-instance, office, guidance or proposal
status
current, ongoing, narrowed or overruled
source_group
For papers: policymaker, academia or professional
primary_url
Official text, where one is linked
source_url
Secondary summary the entry was checked against
reviewed
Whether a lawyer has reviewed the record against the primary source

Corrections

Spotted an error, a missing decision or a better source? Write to hello@havenmarkip.com. Corrections are made in the open and reflected in the data date below.

See it applied: State of play, Where courts agree, Proposals.