Human oversight of AI in investigative and legal work

Position

USCLI holds that justice professionals may use AI and automated tools to assist their work, provided that:

1. A person decides. No decision that affects a person’s rights, liberty, reputation, employment, or professional standing is delegated to an automated system. A named professional reviews the output and makes the decision.
2. The professional understands the tool. Before relying on a tool, the professional understands what it was built to do, the data it draws on, and its known failure modes, and can explain those limits if asked.
3. Outputs are verified. Generated findings, summaries, transcriptions, identifications, and citations are checked against original sources before they appear in a report, filing, or testimony.
4. Use is disclosed where it matters. AI assistance is disclosed whenever law, court rule, or professional standard requires it, and whenever a reader would reasonably want to know.
5. Evidence keeps its integrity. Digitally processed evidence keeps its chain of custody, and the original material is preserved alongside any enhanced, filtered, or generated derivative.

Reasoning

Every discipline USCLI serves produces work that can end up in a courtroom, where it will be tested by someone whose job is to find its weaknesses. Automated tools introduce failure modes that are easy to miss: confident errors, fabricated citations, biased training data, and outputs that cannot be reproduced. The professional who signs the report or takes the stand bears the consequences of those failures, and so do the people the work is about.

Accountability cannot be shared with software. Requiring a named person to decide keeps responsibility where courts, licensing boards, and the public already expect it to be.

Relationship to the Code of Conduct

This statement expands section 6 of the Code of Conduct, Technology and artificial intelligence.

Comment

This statement is open for comment through November 14, 2026, at ethics@uscli.org.