- AI Sanctions
- Citation Verification
- Legal AI
- Duty of Candor
Eyes on the Ball, Kid.
On this page
Attorneys are responsible for what they file. Their AI tools should make verification practical.
What did the New Mexico Supreme Court order?
On September 9, 2026, the New Mexico Supreme Court held an attorney in direct contempt after he admitted filing a ChatGPT-assisted brief containing fabricated witnesses, false testimony and misrepresented legal authority without verification. Its nonprecedential order required a $5,000 payment, referred him for discipline, barred him from appearing before the court pending that process, struck the briefing and ordered replacement counsel. The order also addressed his failure to inform his client. State v. Sandoval; In re Stephen D. Aarons, No. S-1-SC-40845 (N.M. Sept. 9, 2026).
Why isn't model intelligence enough for litigation?
This was ChatGPT, purportedly one of the smartest AIs on the planet, yet it made up not only the law but the facts. Model intelligence alone is not enough for litigation. My unsolicited advice: avoid general-purpose chatbots for litigation. They may be great for transactional work, but they aren't good enough for the courtroom. Choose a platform that makes its sources, its checks and its gaps visible. If it cannot help you verify its work, find a better platform.
What are the two lessons?
Lesson 1: Verification must cover both the law and the factual record.
Lesson 2: Legal AI should always prove itself.
What should "verified" mean in a legal AI tool?
Choose an AI tool that doesn't verify for you, but with you. "Verified" means you are able to put eyes on it.
Did the system allow you to verify the case exists, a quotation is accurate, the authority supports the proposition, a factual assertion appears in the record, or the authority remains good law? Vendors can streamline and supplement this verification, but never replace it.
These are distinct checks. Confirming one does not establish the others.
Why can a brilliant model still be a bad litigation product?
General-purpose chatbots aren't built to let you stand up in court and argue with confidence. Frontier labs build powerful general-purpose AI. That does not make their chatbots good litigation tools. A model can write beautifully while misreading the law, distorting the record or inventing facts. Litigation demands a product built around governing authority, case evidence, procedural posture and practical verification. If the attorney must reconstruct the support for every sentence, the tool has shifted the work into cleanup. A brilliant model can still be a terrible litigation product.
Find a tool that was built for litigation, or you're probably going to end up in one of these blog posts.
Every cite verified against the source. See how Citation Verification works inside the drafting workflow.
See Citation VerificationFrequently asked questions
What happened in State v. Sandoval and In re Aarons?
Does verifying AI output only mean checking citations?
What should verified mean in a legal AI tool?
Are general-purpose chatbots good enough for litigation?
Keep reading
Judge AI: OurFirm.ai's Judicial Intelligence Engine
Judge AI turns years of docket history into a ruling profile you can act on before you file: grant rates by motion type, argument preferences, oral argument patterns, and case-matched precedent for your specific judge.
State of the Practice3 min readThe Review Queue Needs an Admission Policy
When drafting becomes cheap, the scarce resource is the reviewer's willingness to read. A firm that lets every generated alternative into the partner's queue gets slower while every associate gets faster.
Security Review2 min readSecurity Built for Privilege: OurFirm.ai Receives SOC 2 Type 2 Attestation
OurFirm.ai has received its SOC 2 Type 2 attestation from an independent examination of its security controls, tested for design and operating effectiveness over a period rather than at a single point in time.