Judges Use AI, Lawyers Get Sanctioned: Is It a Double Standard?

Here is the 2026 split that has the bar talking. A Northwestern and NYC Bar survey found that most responding federal judges use AI in their own work. Meanwhile courts keep sanctioning lawyers who file AI-fabricated citations. Read fast, it looks like a rule for thee but not for me. The sources are more complicated: judges have also had AI errors in their own orders, and the sanction orders against lawyers focus on filing unverified work and on candor afterward. The common thread is whether anyone checked the output before it counted.

Reviewed and corrected on 12 October 2026. This is general information, not legal advice; check the cited sources before relying on a figure.

TL;DR

• 61.6% of responding federal judges said they use at least one AI tool in their judicial work. This comes from a random-sample survey by Northwestern University with the NYC Bar Association, reported by LawSites on 30 March 2026. Of 502 judges surveyed, 112 responded, a 22.3% response rate, so it describes the judges who answered.
• Use is mostly occasional. Only 5.4% reported daily use and 17% weekly.
• Judges have had AI errors too. In October 2025 two federal judges told Senator Grassley that staff use of AI led to error-ridden opinions that were withdrawn.
• Lawyers are being sanctioned for filing unverified AI output. Examples in 2026 include $15,000 per lawyer in Whiting (6th Cir.), $2,500 in Fletcher (5th Cir.) and $2,500 each plus a six-month suspension in Lnu (9th Cir.), and the $110,204.38 total in Couvrette (D. Or.).
• The orders point to verification and candor. The Ninth Circuit said the rules are violated "at the point of signing and filing," not at the point of research and drafting.
• For in-house teams the takeaway is a workflow rule, not a ban. Check every citation against a primary source before it leaves the building.

What the survey says

Researchers at Northwestern University, working with the New York City Bar Association, ran a random-sample survey of federal judges in late 2025. It was reported by Bob Ambrogi on LawSites on 30 March 2026 and is forthcoming in volume 27 of The Sedona Conference Journal (LawSites). The details that matter:

• It surveyed 502 bankruptcy, magistrate, district and appeals judges. 112 responded, which is a 22.3% response rate.
• 61.6% of responding judges use at least one AI tool in their judicial work. 5.4% use one daily, 17% weekly, and 19.6% monthly. 38.4% reported never using any of the listed tools.
• Westlaw AI-Assisted Research or Deep Research was the most commonly used tool (38.4% reported some use), and ChatGPT was second at 28.6%.
• The most common use was legal research, reported by 30% of judges. For chambers staff, legal research was the top use at 39.8%.
• 1.8% of judges said they use AI to "make decisions" and 4.5% to "inform decisions." Only 1.8% reported using it to draft filed documents such as orders or opinions.
• Chambers policies were mixed: 7.4% permit and encourage, 25.9% permit, 20.4% formally prohibit, 17.6% discourage, and 24.1% have no official policy.

One judge told the researchers that a clerk wrote a memo and then asked AI to write one on the same question. "Of the 11 cases AI cited in its version, 10 of them were fake."

Judges have had AI errors too

In October 2025, two federal judges answered questions from Senate Judiciary Chairman Charles Grassley. According to the ABA Journal, District Judge Henry T. Wingate (S.D. Miss.) said a law clerk used Perplexity to synthesize information on the docket, and that his 20 July opinion, a temporary restraining order, referred to nonexistent allegations, parties and declarations. District Judge Julien Xavier Neals (D.N.J.) said a law school intern used ChatGPT for legal research without authorization, and his 30 June opinion misstated case outcomes and used fake quotes. Both said the opinions were early drafts that should not have been docketed, and both withdrew them (ABA Journal, 24 October 2025).

What the sanction orders say about lawyers

Courts have responded to lawyers' filings with fines, fee awards, suspensions and referrals. The ones most often cited in 2026:

• Couvrette v. Wisnovsky (D. Or., orders of 12 December 2025 and 23 March 2026). Fifteen non-existent cases and fabricated quotations attributed to eight authorities. The court struck the briefs, dismissed the claims with prejudice, ordered a $15,500 penalty on one lawyer, and awarded $94,704.38 in fees, for a total of $110,204.38 across two lawyers (December order, March order). The court said that deleting the fake cases but leaving the statements of law was no correction and called it a coverup. See the Oregon sanction breakdown.
• Whiting v. City of Athens (6th Cir., 13 March 2026). Over two dozen fake citations and misrepresentations of fact. Full appellate fees and double costs under Rule 38, plus $15,000 from each of two lawyers as punitive sanctions under the court's inherent authority. The court said "smaller fines have plainly been inadequate" (opinion).
• Fletcher v. Experian (5th Cir., 18 February 2026). $2,500 for a reply brief with 16 fabricated quotations and five other serious misstatements (opinion).
• Lnu v. Blanche (9th Cir., 3 June 2026). $2,500 each and a six-month suspension from practice before the court (opinion).

For more rows, see the sanctions tracker and our sanctions guide. We have not verified any aggregate dollar total for a quarter. One trade-press tally put first-quarter 2026 sanctions at $145,000 but used rounded and different figures than the orders, so we do not repeat it.

Is it a double standard?

Not in the sense of different rules for the same conduct. Here is what the sources support.

Different consequences. A lawyer who files a brief with fake authority faces Rule 11, appellate rules, inherent authority and bar discipline. The judges' withdrawn orders were handled by withdrawal and docket notices that the opinions had been removed. Judge Wingate wrote that he had "taken steps in my chambers to ensure this mistake will not happen again," and Judge Neals said the standard practice in his chambers is for draft opinions to go through several layers of review.

The same underlying risk. In the judges' letters and in the sanction orders, the failure was that unchecked AI output made it into a document that left the building. The Ninth Circuit put the lawyer's version this way: the rules "are not violated at the point of research and drafting, but at the point of signing and filing."

Candor matters in the lawyer cases. Courts in Fletcher and Lnu said lesser sanctions were likely or might have been warranted if counsel had been forthcoming. In Couvrette the court found the cosmetic fix was a coverup. We do not read these orders as a formula for penalty size.

The rule was never "do not use AI"

ABA Formal Opinion 512 (29 July 2024) applies existing duties to generative AI. It says relying on or submitting a tool's output without "an appropriate degree of independent verification or review" could violate Rule 1.1, and that the right level "will necessarily depend on the GAI tool and the specific task that it performs." It also says output used in court must be "carefully reviewed to ensure that the assertions made to the court are not false." We walk through the opinion in our ABA Formal Opinion 512 guide.

What in-house counsel should do

An unchecked AI citation in a demand letter, a compliance memo or a board deck is a credibility problem, and in regulated work it can be a liability problem. A ban is one option, but the orders point to a workflow answer:

1. Make verification a step. Every AI-supplied authority is opened in a primary source and read before it leaves the team. See how to verify AI legal citations before filing.
2. Write the policy down. In the survey, 24.1% of judges said they had no official policy. 512 expects managerial lawyers to "establish clear policies regarding the law firm's permissible use of GAI." Cover approved tools, what must be human-reviewed, and what gets disclosed.
3. Know what happens to inputs. Before pasting client information into a tool, check its terms. See where your legal AI data actually goes.

FAQ

Do federal judges actually use AI?

According to a random-sample survey by Northwestern University and the NYC Bar Association, reported in March 2026, 61.6% of responding judges use at least one AI tool. 112 of 502 judges responded. Only 5.4% use one daily and 17% weekly.

Why do judges get to use AI when lawyers get sanctioned?

Lawyers are sanctioned for filings with fake or misrepresented authority and for how they responded when it came to light, not for using AI. The Ninth Circuit said so in Lnu. Judges have also had AI errors: in October 2025 two judges said staff use of AI produced error-ridden orders that they withdrew.

How much have lawyers been sanctioned for AI hallucinations?

Among the orders discussed in our sanctions guide, amounts range from $1,000 (a signing lawyer in Wadsworth) to $110,204.38 in total. The latter is the total in Couvrette v. Wisnovsky (D. Or.): a $15,500 penalty and $94,704.38 in fees across two lawyers. Appeals courts have fined $2,500 (Fletcher), $15,000 per lawyer (Whiting) and $2,500 each plus a six-month suspension (Lnu).

What is ABA Formal Opinion 512?

It is the ABA's formal ethics opinion on generative AI, issued 29 July 2024. It applies the existing Model Rules, including competence, confidentiality, communication, candor, supervision and fees. It does not ban AI and it does not set a fixed verification standard.

Can I get sanctioned for using AI if I check the citations?

The orders sanction filing unverified authority and failures of candor, not AI use. A lawyer who opens each case, confirms the quotes and reads the holding is doing what the courts say is required.

What is the safest way for an in-house team to use AI?

Write a policy that names approved tools and required review, check every citation and quotation against a primary source before it leaves the team, and answer accurately if anyone asks how a document was prepared.

For the US statutes, regulations and court rules these laws sit in, see the US primary law API.

Arshita Anand

Co-Founder & CEO · Attorney

Arshita leads product and strategy at Vaquill AI, the US primary law data API. As an attorney, she decides which law we carry, how each section is sourced and cited, and what legal AI and compliance teams need from the data.