AI in lawmaking

The drafting office decides.

Attorneys and drafters author the law. Software should not. AI has a real role to play, taking on slow, mechanical work and leaving drafters more time for the judgment only they can provide, but the law is too important to hand over to a tool that can’t be held accountable.

My six commitments for AI in legislative drafting

Read the full essay →

This is what I think a drafting office should expect from any software that brings AI anywhere near the law.

  1. 1The drafting office stays in chargeAI should return proposals, never edits. The office’s drafters accept, change or reject each one, and nothing reaches the text without that decision.
  2. 2The deterministic core never requires AIEverything needed to draft, amend, compare and publish law should work with AI switched off.
  3. 3Confidentiality comes firstDrafts are confidential until they are introduced. Insist on zero data retention, and never let your material train a model without written consent.
  4. 4Processing stays within your boundariesBy default, AI should run where you say it runs. You may choose otherwise; your vendor never should.
  5. 5No dependence on any one AI vendorModels should sit behind a single gateway, so they can be replaced without rework.
  6. 6Authoritative data firstAI should answer from your own official text, and every proposal should cite a provision and quote the words it relied on.

The drafting office decides. Always.— Grant Vergottini

Xcential has adopted these commitments for its own drafting software, but they aren’t ours to own. I’d like to see every vendor in this field, and every drafting office buying from one, hold the line on them.

Three arguments

All AI essays →

Published work

All publications →