Skip to main content
An AI use rule is a rule in which a court or judge says something about artificial intelligence (AI) in filings or proceedings. Some ask something of the filer, such as saying that AI was used. Others only remind filers of duties they already have, or set a court’s policy for its own staff. The API marks these rules with a topics field and finds them with topic=ai_use.

Why teams use it

  • Check a judge before filing. Read a judge’s rules and look for a rule whose kind is requirement: a disclosure, a certificate, a ban on a use of AI.
  • Keep a list of orders current. Page through topic=ai_use for a court and keep the rules whose kind is requirement.
  • Show the court’s own words. Each rule keeps the court’s quote and the address of the court’s document, so a filer can read the order itself.
  • Tell a duty from a mention. kind separates an order that asks something of the filer from a reminder, a passing mention and a court’s policy for its own staff.

How it works

  • Every rule the API returns can carry a topics field. A rule about AI use has topics.ai_use. A rule about anything else has no topics field.
  • topics.ai_use has three parts. kind says what the rule is. label says what it is mainly about. fields answers specific questions, such as whether the rule asks for a disclosure, who must follow it and how to comply.
  • A rule has the topic once the document it comes from has been read for it. A document that has not been read yet has no AI use rules in the API, so an empty answer does not prove that a court has no AI order.
  • ai_use is the only topic today. topics is keyed by topic, so read the keys you know and ignore the others.

Terms

Kinds

Every AI use rule has one kind. When more than one could apply, the first of these wins, in the order court_internal, reminder_only, requirement, other. A rule that only reminds filers of an existing duty, such as the duty to check their own work, is a reminder_only even when one of its yes or no fields says yes.

Labels

Fields

A yes or no field holds yes, no or unknown. yes means the rule says the thing described and no means it does not. A field that lists values holds one of them or unknown. not_stated and none_stated are confirmed answers: the rule says nothing on the point.

Find the rules

The examples use a made-up court, judge and order.

1. Search a court

GET /api/v1/extracted-rules with topic=ai_use returns only the rules that have the topic. Add district_id for one court. The other filters of that endpoint work with it.
meta.total counts the rules that have the topic, and paging works as it does for any search. Each rule is one row against your daily limit, however much its topics holds.

2. Read a judge’s rules

GET /api/v1/rules returns a judge’s rules in two layers, the court’s and the judge’s own, and each rule carries its topics when it has them. Read kind first, then fields for the detail.

3. Keep the orders

Keep the rules that ask something of the filer, and show each one with the court’s quote and its source.

Behavior details

  • The filter returns every kind. topic=ai_use returns reminders, mentions and a court’s own policies as well as orders. There is no filter by kind yet, so check topics.ai_use.kind in each result.
  • The filter runs before the count and the paging. meta.total and the pages are those of the rules that have the topic.
  • It combines with the other filters. district_id, judge_slug, include_court_rules, logic_type, workflow_phase, case_type and q narrow the rules that have the topic.
  • A rule without the topic has no topics. The field is left out, not empty. A rule that was read and found not to be about AI use looks the same as a rule that was not read.
  • Court-wide and judge rules both carry topics. A court-wide rule has judge_slug court. Read both layers: a court-wide order applies to every courtroom of the court.
  • Rulebook requirements carry topics too. A requirement read from a rulebook shows topics when it has them, but topic does not filter them. Sending rulebook_id with topic is a 400.
  • Compliance profiles have no topics. /rules and /profile return a compliance profile for a court that has one (edny today), and a profile carries no topics. Read that court’s AI use rules with /extracted-rules?topic=ai_use&district_id=edny.
  • Order. Rules come newest first, as they do for every search of /extracted-rules.

Errors and limits

400
  • A rule appears once its document was read. A court’s new order is not an AI use rule in the API until its document has been read for the topic.
  • unknown is not an answer. It means the answer was not confirmed, so do not read it as no.
  • Read the court’s document before filing. The API reports what a court’s document says, and source_url links to it. It does not work out which rule governs a filing or whether a rule applies to your case.
  • A text or case-type search scans at most 10,000 rules. With topic the scan covers only the rules that have the topic.

Common mistakes

  • Counting every result as an order. Results include reminders, mentions and a court’s own policies. Count the rules whose kind is requirement.
  • Reading topics.ai_use.label as the duty. The label names the subject. A rule labeled verification can still be a requirement, and a rule labeled disclosure can be a reminder_only.
  • Treating unknown as no. unknown means the answer was not confirmed, and no means the rule does not say it.
  • Reading a missing topics as “no AI rule”. The rule may not be about AI use, or its document may not have been read for the topic.
  • Sending topic with rulebook_id. Rulebook requirements cannot be filtered by topic yet.