When AI Lies to the Court 

What Washington Attorneys Need to Know Now

A gray robot on a bright pink background. The robot is labeled "AI"
Illustration © Getty/Overearth

BY GLORIA DOMINGOS AND DR. DARIA LUNDELL (KOUCHERETS) 

In August 2025, U.S. Magistrate Judge Alison Bachus of the District of Arizona sanctioned a Washington attorney for submitting a brief riddled with fabricated, misleading, and unsupported citations.11 Mavy v. Comm’r of Soc. Sec. Admin., No. CV-25-00689-PHX-KML (ASB) (D. Ariz. Aug. 14, 2025). Then she did something unusual. She ordered the court clerk to send the sanctions order directly to the Washington State Bar Association “so that the bar association may take any action it deems appropriate.”22 Id. at 24. She also revoked the attorney’s pro hac vice status33 Id. at 19. and required her to personally write to three federal judges whose names AI had falsely inserted into invented opinions.44 Id. at 23. Twelve of 19 citations in that brief were fabricated. The judges were real. The cases were not.55 Id. at 9.

After almost a year, Washington still has no binding rule exclusively addressing AI.

That was not the only message to arrive. In Oregon, U.S. Magistrate Judge Mark D. Clarke dismissed a case and imposed a $110,000 penalty against two lawyers after they submitted briefs containing citations and quotations hallucinated by AI.66 Couvrette v. Wisnovsky et al., No. 1:21-cv-00157-CL (D. Or. 2026). In Colorado, attorney Zachariah Crabill lost his license for 90 days after submitting ChatGPT-fabricated citations.77 People v. Crabill, No. 23PDJ011 (Colo. O.P.D.J. Nov. 22, 2023). When the judge asked about them at a hearing, Crabill blamed a legal intern. Days later he filed an affidavit admitting he had done it all himself and that “like an idiot” he had not checked. The denial made it worse.

In July 2025, three experienced Butler Snow LLP attorneys—a firm with over 400 lawyers and a written prohibition on unverified AI—were disqualified from a case in the Northern District of Alabama for submitting five fabricated citations.88 Johnson v. Dunn, No. 2:21-cv-01701-AMM, 2025 WL 2086116, at *19–20 (N.D. Ala. July 23, 2025). Their seniority was no protection. Judge Anna M. Manasco sent notifications to bar regulators in every state where they held licenses. She was not subtle about why: High status in a firm does not reduce your verification obligation; it increases it.

Then there are the judges themselves. Federal Judge Julien Xavier Neals of New Jersey withdrew his own ruling in In re CorMedix Inc. Securities Litigation after a law school intern used ChatGPT to draft the research.99 In re CorMedix Inc. Sec. Litig., No. 2:21-cv-14020 (D.N.J. 2025). Federal Judge Henry T. Wingate of Mississippi vacated an order in a case involving enforcement of a state ban on DEI programs in public schools after his clerk used Perplexity AI—producing fabricated party names, invented precedents, and wrong statements of state law.1010 Jackson Fed’n of Teachers v. State, No. 3:25-cv-00417 (S.D. Miss. July 23, 2025). Wingate called it a “technical mistake” and tried to remove the original from the public docket. Sen. Chuck Grassley (R-IA) opened a formal Senate inquiry and demanded the documents back.1111 Letter from Sen. Chuck Grassley, Ranking Member, Senate Judiciary Comm., to John G. Roberts, Jr., Chief Justice of the United States (Oct. 6, 2025), available at www.grassley.senate.gov/download/grassley-to-the-honorable-henry-t-wingate_-ai-oversight.

French researcher Damien Charlotin, who tracks AI hallucination cases globally, was logging two incidents per week by spring 2025.1212 www.damiencharlotin.com/hallucinations/. As one Illinois court publication put it: “paste in haste, repent at leisure.”1313 www.illinoiscourts.gov/News/1665/Paste-in-Haste-The-Fallout-of-AI-Hallucinations-in-Court-Filings-and-the-New-ARDCs-Guide-to-Implementing-AI/news-detail/.

Washington’s response to all of this? An advisory opinion.

We practice immigration law at Domingos Law Firm in SeaTac. Our clients speak more than 10 languages: Ukrainian, Russian, French, Portuguese, Spanish, Wolof, and several West and Central African dialects. We use AI. The volume demands it.

But AI hallucinations in immigration practice do not look like fabricated federal citations. They look like a misspelled village name in Guinea-Bissau. A nonexistent regional official in Senegal. A government decree in Angola that never existed. Westlaw will not catch these. Westlaw does not index village-level political geography in West Africa.

When an asylum seeker’s case turns on whether a specific official held power in a specific region in a specific year, a hallucinated detail is not a technical error. It can be the difference between protection and deportation.

At our law firm, we implemented internal guidance requiring verification of all materials drafted with the help of AI. Not because Washington requires it, but because our clients cannot afford for us not to.

That needs to change. While implementing rules at individual law firms is important, there should still be general guidance applying to all practitioners in the state.

Before we get to what we believe Washington should do, let’s look at what the courts have already decided.

In Johnson v. Dunn, the fabricated citation was inserted by a practice group leader. Not an intern. Not a paralegal. A senior partner at a national firm. The firm had a written AI policy. He ignored it. The court rejected every excuse offered: that a supervisor made the error, that the hallucinated citation happened to support a correct legal point—“a stroke of pure luck,” the court said drily—and that the attorney thought the citation came from prior reliable filings. None of it mattered. Your signature means you are responsible for every word. Full stop.

Three Washington rules already say this. RPC 1.1 requires technological competence, which today means knowing that leading AI legal research tools hallucinate in 17 to 33 percent of queries, per a 2024 Stanford study. If you did not know that, you are already behind. RPC 3.3 prohibits knowingly submitting false statements to a court—and once opposing counsel raises a question about your citations, defending them instead of disclosing the problem is its own violation, as Mata v. Avianca (S.D.N.Y. 2023) made clear. RPC 1.6 covers confidentiality—and entering a client’s account of persecution into a free AI platform that retains user data for model training may already be a violation, before a single citation is ever filed.

WSBA Advisory Opinion 2025-05 confirmed all three rules apply. Washington attorneys can be disciplined under these rules if they skip citation verification and consequently include false or misleading information in a court filing. But we argue that these rules don’t go far enough. They do not specifically address AI, and they do not state exactly what verification and disclosure require.

In a survey of WSBA members conducted by the National Business Research Institute in 2024, 77 percent of respondents said they did not believe that Washington’s current ethical rules and guidelines adequately covered the use of generative AI.1414 Find all materials related to the WSBA’s Legal Technology Task Force at www.wsba.org/connect-serve/committees-boards-other-groups/legal-technology-task-force. Seventy-two percent of respondents said they supported adding an MCLE requirement for a technology-related CLE. When asked about personal knowledge around generative AI, only 9 percent of respondents rated themselves as “Good” or “Better.” Thirty-six percent said their knowledge on the topic was “Fair” and 28 percent said “Poor.” 

Washington lawyers want additional knowledge and guidance in a rapidly evolving domain. 

We propose a binding AI Verification Standard under RPC 1.1 built around actual risk. Not all AI use is the same, and the rules should not treat it that way.

Routine tasks—formatting, grammar, scheduling, administrative work—need basic review before they go out. The same standard as paralegal work under RPC 5.3. Nothing more.

Substantive drafting—motions, briefs, contracts, declarations without case citations—needs real attorney review. Read every factual claim. Check every statement of law. AI writes from training data, not from what the law says today. It will be confident. It will sometimes be wrong.

Legal research and citations need the strictest standard. Every citation, every time, verified against Westlaw, LexisNexis, or official court records before a document goes to a tribunal. No exceptions. And do not ask the AI to check its own work. That is exactly what happened in Mata v. Avianca.1515 Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023). The attorney asked ChatGPT to confirm the citations. ChatGPT confirmed them. Judge Castel ordered the attorney and his partner to each pay $5,000 in fines.1616 www.courthousenews.com/sanctions-ordered-for-lawyers-who-relied-on-chatgpt-artificial-intelligence-to-prepare-court-brief/.

For immigration work, verification goes further. Country conditions research must be cross-referenced against U.S. Department of State reports, UNHCR documentation, and human rights records—because AI invents officials, decrees, and histories in places most databases have never indexed. The attorneys most likely to face this serve the communities least able to absorb the consequences of an error.

One more thing: disclosure. At Domingos Law Firm we tell our clients when we use AI. Our clients have fled governments that lied to them. They deserve to know how their legal work is produced. Washington should require a simple certification that AI-generated citations have been independently verified before any court submission. Courts already require attorneys to certify accuracy. This is the same obligation, updated for 2026.



A federal judge sent a sanctions order directly to the WSBA. Two attorneys in Oregon had their case dismissed and received massive fines. A senior partner at a 400-lawyer firm learned that seniority means nothing when your signature is on a hallucinated brief. A Colorado attorney blamed an intern, then admitted it was him. Two federal judges withdrew their own rulings. A United States Senator opened a formal inquiry.

This is not a problem for time-crunched solos or overwhelmed associates. It has reached practice group leaders, law clerks, and the judges themselves.

Washington attorneys are good lawyers. Many of them are asking for guidance. They deserve an answer.

The WSBA has the tools. The courts have delivered the message—by name, directly to this bar association. The response is overdue. 

Headshot of Gloria Domingos

About the authorS

Gloria Domingos is the founding attorney and CEO of Domingos Law Firm in SeaTac. 

Headshot of Daria Lundell Koucherets

Daria Lundell  (Koucherets), Ph.D., is COO of Domingos Law Firm. 

Together they are co-authors of Lex AI: A Global Legal Language. The firm serves clients from more than ten language communities across immigration, asylum, and family-based matters.

NOTES

1. Mavy v. Comm’r of Soc. Sec. Admin., No. CV-25-00689-PHX-KML (ASB) (D. Ariz. Aug. 14, 2025).

2. Id. at 24.

3. Id. at 19.

4. Id. at 23.

5. Id. at 9.

6. Couvrette v. Wisnovsky et al., No. 1:21-cv-00157-CL (D. Or. 2026).

7. People v. Crabill, No. 23PDJ011 (Colo. O.P.D.J. Nov. 22, 2023).

8. Johnson v. Dunn, No. 2:21-cv-01701-AMM, 2025 WL 2086116, at *19–20 (N.D. Ala. July 23, 2025).

9. In re CorMedix Inc. Sec. Litig., No. 2:21-cv-14020 (D.N.J. 2025).

10. Jackson Fed’n of Teachers v. State, No. 3:25-cv-00417 (S.D. Miss. July 23, 2025).

11. Letter from Sen. Chuck Grassley, Ranking Member, Senate Judiciary Comm., to John G. Roberts, Jr., Chief Justice of the United States (Oct. 6, 2025), available at www.grassley.senate.gov/download/grassley-to-the-honorable-henry-t-wingate_-ai-oversight.

12. www.damiencharlotin.com/hallucinations/.

13. www.illinoiscourts.gov/News/1665/Paste-in-Haste-The-Fallout-of-AI-Hallucinations-in-Court-Filings-and-the-New-ARDCs-Guide-to-Implementing-AI/news-detail/.

14. Find all materials related to the WSBA’s Legal Technology Task Force at www.wsba.org/connect-serve/committees-boards-other-groups/legal-technology-task-force.

15. Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023).

16. www.courthousenews.com/sanctions-ordered-for-lawyers-who-relied-on-chatgpt-artificial-intelligence-to-prepare-court-brief/.