Net Good IndexSubmit a correction

Harm Ledger · verified · Computer Science

New Mexico Supreme Court holds a lawyer in contempt for ChatGPT-fabricated witnesses

In September 2026 the New Mexico Supreme Court held attorney Stephen Aarons in contempt, fined him $5,000, and referred him for discipline after a murder-appeal brief prepared with ChatGPT included wholly fabricated witnesses and false police testimony.

9 Sep 2026Tier 2 Significant HarmMethodology 0.1

Current score

0.24

3 base · Significant Harm (tier 2 of 5, 3 pts)
× 0.3500 attribution · Material acceleration
× 1.0000 evidence · Overwhelming validation or measured outcome
× 1.0000 realization · Realized outcome
× 0.4500 durability
Event-level product before credit split: 0.47

Fabricated witness testimony in a murder appeal is significant institutional harm (tier 2). Lawyers had a duty to check (0.35). The contempt order is a primary court record (1.0). Realization is the filed brief and sanction. Durability is the record and discipline referral, not a population injury.

What happened

Aarons told the court he fed transcripts and case materials to ChatGPT expecting a reliable summary and did not verify the brief. The court found fictitious officers and witnesses, false testimony about clothing and threats, and misrepresented authorities in a live murder appeal. This is a documented integrity failure in a criminal appellate proceeding, worse in kind than typical hallucinated citations, but still case-level rather than population harm.

Model attribution

0.24

GPT

Generated fabricated witnesses, police testimony, and misrepresented authorities that were filed in the brief in chief.

ChatGPT produced the false material; counsel filed it without verification.

Attribution 0.3500 · Credit share 50% · OpenAI

Claims

  • The New Mexico Supreme Court found that a ChatGPT-assisted murder-appeal brief contained false testimony from wholly fabricated witnesses and held counsel in contempt.

    outcome · supported

Sources

primary sources

independent sources

Revision history

  • 13 Sep 2026 · 0.00 0.24

    Imported events under methodology 0.1.