AI Hallucination Sanctions: Lawyer Cases and How to Avoid Them

The four independent checks every citation passes before it reaches your draft.

AI Hallucination Sanctions, in One Line

Courts sanction lawyers when they file AI-generated briefs that cite cases, quotes, or holdings the AI invented and the lawyer never checked. Penalties run from a $5,000 fine in Mata v. Avianca to a $110,000 award in a December 2025 Oregon case, with a two-year license suspension in between. By early 2026 the problem reached the federal appeals courts: the Fifth, Sixth, and Ninth Circuits all handed down sanctions, and U.S. courts imposed more than $145,000 in AI hallucination penalties in the first quarter alone (ComplianceHub, 2026). The fix is not avoiding AI. It is verifying every citation against a primary source before you file.

TL;DR

• Lawyers keep getting sanctioned for filing AI-generated briefs with fabricated citations. Mata v. Avianca ($5,000), Crabill (two-year suspension), Wadsworth v. Walmart ($3,000 + $1,000 + $1,000), a $10,000 California appellate fine, a $31,100 K&L Gates sanction, and a $110,000 Oregon penalty are the canonical entries.
• The appeals courts have joined in. In 2026 the Fifth Circuit ($2,500), the Sixth Circuit ($15,000 per lawyer under Rule 38), and the Ninth Circuit ($2,500 each plus a six-month suspension) all sanctioned attorneys for AI-fabricated authority. This is no longer a trial-court curiosity.
• The numbers are climbing fast. Damien Charlotin's hallucination tracker logged 1,624 cases worldwide as of 18 June 2026, up from a few per month in early 2025 to several per day now. U.S. courts alone crossed $145,000 in sanctions in Q1 2026 (ComplianceHub, 2026).
• General-purpose chatbots are language models, not retrieval systems. Stanford HAI measured hallucination rates of 58 to 88% on raw LLMs and 17 to 33% on purpose-built legal AI. Even the best legal RAG tools need verification.
• ABA Formal Opinion 512 (July 2024) is the operating manual: competence, confidentiality, candor, supervision, fees. There is no "the AI did it" defense.
• Verification is a workflow. The seven-step protocol below makes the duty checkable rather than aspirational.

Part of our legal AI verification and hallucination guide series.

For related verification / hallucination / vendor-trust coverage, see How to Verify AI Legal Citations Before You File (ABA 512 Checklist) and Damien Charlotin's Hallucination Tracker, Read Like a Risk Manager.

In June 2023, Judge P. Kevin Castel of the Southern District of New York fined lawyers Steven Schwartz and Peter LoDuca $5,000 for submitting a legal brief packed with fabricated case citations. The case was Mata v. Avianca, Inc. (1:22-cv-01461, S.D.N.Y.), a straightforward personal injury claim.

The brief cited six cases that did not exist. Not one of them appeared in Westlaw, Lexis, or any other legal database, because they were invented wholesale by ChatGPT.

What made the situation worse: when opposing counsel flagged the bogus citations, Schwartz went back to ChatGPT and asked it to confirm the cases were real. The chatbot obliged, insisting the cases "can be found in reputable legal databases such as LexisNexis and Westlaw."

Schwartz submitted that assurance to the court. Judge Castel found that the lawyers acted in "subjective bad faith" under Federal Rule of Civil Procedure Rule 11.

That was 2023. The problem has gotten worse since then.

This Keeps Happening

Mata v. Avianca was the first high-profile AI hallucination sanctions case. It was not the last.

A growing database maintained by legal researcher Damien Charlotin counted 1,624 known cases worldwide as of 18 June 2026 (AI Hallucination Cases, damiencharlotin.com). When tracking started in early 2025 it logged a few cases a month. By 2026 it was logging several a day, and a large share involve licensed attorneys rather than self-represented litigants.

Here are the cases that followed, in order of how hard they hit:

Crabill, Colorado (November 2023). Lawyer Zachariah Crabill used ChatGPT to find supporting case law for a motion to set aside a summary judgment ruling in El Paso County District Court. He had never drafted that type of motion before.

The morning of the hearing, he realized the citations were fabricated, texting his paralegal that the ChatGPT case cites were "garbage." When the judge asked about the errors in open court, Crabill panicked and blamed a legal intern.

The Colorado Office of Attorney Regulation Counsel brought disciplinary proceedings. Crabill received a two-year suspension with 90 days actively served.

Wadsworth v. Walmart (D. Wyo., February 2025). Three lawyers, including two from Morgan and Morgan (the 42nd largest firm in the U.S. by headcount), filed a motion in limine that contained eight citations to cases that did not exist.

Lawyer Rudwin Ayala had used the firm's in-house AI platform, MX2.law, prompting it to "add Federal Case law from Wyoming setting forth requirements for motions in limine." He pasted the output into the brief without checking a single citation.

Judge Kelly H. Rankin revoked Ayala's pro hac vice status and fined him $3,000. Co-counsel T. Michael Morgan and Taly Goody were each fined $1,000. All three signed the brief, and all three were sanctioned. The two who did not draft it were still on the hook because they signed it.

California Appeal, Mostafavi (September 2025). Los Angeles lawyer Amir Mostafavi was fined $10,000 by California's Second District Court of Appeal for filing an appellate brief in which 21 of 23 case quotations were fabricated by ChatGPT (CalMatters, September 2025). The quoted holdings themselves were invented, not only the citations.

Lacey v. State Farm, K&L Gates (May 2025). A special master sanctioned K&L Gates (a 1,700-lawyer firm) and co-counsel Ellis George $31,100 for a brief built on "bogus AI-generated research" in the Central District of California. Retired Magistrate Judge Michael R. Wilner called it a "collective debacle" (ABA Journal, May 2025).

Fifth Circuit (2026). The U.S. Court of Appeals for the Fifth Circuit sanctioned an attorney $2,500 who admitted to using vLex and Thomson Reuters CoCounsel to draft arguments that cited fabricated authority. This is one of the first sanctions tied to purpose-built legal AI tools, not a general chatbot (ABA Journal, April 2026).

Sixth Circuit, Whiting v. City of Athens (March 2026). The Sixth Circuit sanctioned two Tennessee attorneys whose briefs contained more than 24 fabricated citations. Under Federal Rule of Appellate Procedure 38, the panel ordered each attorney to pay $15,000 in punitive sanctions ($30,000 combined), plus double costs and the appellees' attorney fees. The court called that "the stiffest penalty available under Rule 38" and said it wanted to send "the loudest message" that "this type of conduct is not allowed in our court or any other" (Whiting v. City of Athens, No. 24-5918/5919, 25-5424 (6th Cir. 2026); Sixth Circuit Appellate Blog, 2026).

Ninth Circuit, Lnu v. Blanche (June 2026). The Ninth Circuit sanctioned Orange County attorneys Mike Sethi and William Rounds $2,500 each and suspended both from practice before the court for six months. Their briefs cited fabricated cases and false quotes. They first blamed typos, then admitted an unlicensed law school graduate at the firm had used unauthorized AI tools. The court wrote that "lawyers using generative AI must thus be aware of the tendency of generative AI to make these mistakes and guard against them" (Lnu v. Blanche, No. 24-4790 (9th Cir. June 3, 2026); Bloomberg Law, June 2026).

Oregon, Brigandi and Murphy (December 2025). San Diego attorney Stephen Brigandi and Portland attorney Tim Murphy were sanctioned a combined $110,000 (the largest AI hallucination penalty in U.S. legal history) after filing briefs with 15 nonexistent cases and eight fabricated quotations. Magistrate Judge Mark D. Clarke of the District of Oregon called the case "a notorious outlier in both degree and volume" in his December 12 opinion (ABA Journal, 2026). See our breakdown of the Oregon $110,000 sanction for what went wrong step by step.

The pattern across all of these cases is identical. A lawyer uses an AI tool for legal research. The tool generates plausible-sounding case names, citations, and holdings.

The lawyer does not verify any of it. The fabrications make it into a court filing. A judge discovers the fraud.

How a fabricated citation escalates to a sanction

The sanction rarely tracks the hallucination alone. What courts punish hardest is what the lawyer does after the fake cite is caught. Owning the mistake, withdrawing the filing, and reimbursing opposing counsel keeps the penalty low. Blaming a paralegal, defying a show-cause order, or asking a second chatbot to "confirm" the fabrications (as Schwartz did in Mata) escalates it into stacked Rule 11 sanctions, contempt findings, and a bar referral.

In April 2026 the Nebraska Supreme Court indefinitely suspended an attorney's license over AI-fabricated citations, and the Whiting panel reached the Rule 38 ceiling in part because one attorney had prior discipline for lack of candor and defied the show-cause order (ComplianceHub, 2026). The cover-up, not the chatbot, is what turns a fine into a career event.

How Big Is the Hallucination Problem?

Stanford's Human-Centered Artificial Intelligence institute (HAI) has run the most rigorous studies on legal AI accuracy to date.

Their first major study, "Large Legal Fictions," tested general-purpose models on over 800,000 verifiable legal questions. The hallucination rates were staggering: GPT-3.5 hallucinated 69% of the time, GPT-4 came in at 58%, and Llama 2 at 88%.

These are not occasional slip-ups. These models fabricated legal information more often than they got it right.

A follow-up study, "Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools," examined purpose-built legal research platforms. The results were better, but still alarming.

Using data from May 2024 testing, Lexis+ AI hallucinated at a 17% rate. Westlaw AI-Assisted Research hallucinated at 33%. Even as a baseline comparison, GPT-4 with no legal retrieval layer came in at 43%.

The takeaway: retrieval-augmented generation (RAG) significantly reduces hallucinations compared to raw LLM output, but even the best-known legal AI tools still produce incorrect information roughly one out of every five to six queries. That is far too high to trust without verification.

Thomson Reuters reached the same conclusion from the courtroom side. A Westlaw review found non-existent legal authorities in 22 separate cases over a single study window, and in one pro se filing, Powhatan County School Board v. Skinger (E.D. Va.), a single brief carried 42 fabricated citations (Thomson Reuters Institute, 2026).

And the hallucinations are not always obvious. Some are outright fabrications (invented case names, fake docket numbers). Others are subtler: a real case name paired with a fabricated holding, or a correct citation linked to a legal principle the case never addressed.

The subtle ones are more dangerous because they are harder to catch.

Why Legal AI Hallucinates

Understanding why this happens is useful, because it helps you evaluate which tools are safer and which are not.

Large language models generate text by predicting the most probable next word in a sequence. They have no internal database of cases. They have no concept of "truth."

When you ask GPT-4 to cite a case supporting a legal proposition, the model generates text that looks like a case citation because it has seen millions of case citations in its training data. It knows the format. It knows what plausible party names look like. It knows citation conventions.

But it is pattern-matching, not looking anything up.

This is the core problem: LLMs are language generators, not information retrieval systems.

Retrieval-augmented generation (RAG) attempts to fix this by adding a retrieval step. Before the LLM generates its response, the system searches an actual database of cases, retrieves relevant documents, and feeds them to the model as context. The model then generates its answer based on those retrieved documents rather than its training data alone.

RAG helps substantially, but it introduces its own failure modes:

Retrieval misses. The search step might not find the most relevant cases, especially for niche or novel legal questions. If the retrieval layer returns weak results, the LLM may fill the gaps with fabricated material.

Synthesis errors. Even when the correct cases are retrieved, the LLM can mischaracterize them. It might attribute a holding from Case A to Case B. It might invert a legal rule. It might state that a court "held" something that was actually dicta.

Overconfidence. LLMs do not signal uncertainty. A fabricated citation is presented with the same confidence as a verified one. There is no built-in mechanism to flag "I am less sure about this claim."

Context window limitations. When multiple long documents are fed into the model's context, relevant details from earlier in the context can be lost or confused, especially in complex multi-issue research tasks.

These are not theoretical problems. They are the specific mechanisms behind every sanctions case discussed above.

What the ABA Says: Formal Opinion 512

On July 29, 2024, the ABA Standing Committee on Ethics and Professional Responsibility released Formal Opinion 512, its first formal ethics guidance on lawyers' use of generative AI.

The opinion does not ban AI use. It frames AI as a tool that triggers existing ethical obligations. It addresses six areas, all grounded in the Model Rules of Professional Conduct.

Competence (Model Rule 1.1). Lawyers must have "a reasonable and current understanding of the specific capabilities and limitations" of any AI tool they use. This means you cannot claim ignorance of hallucination risks.

If you use an AI tool for legal research, you are expected to know that it can fabricate citations, and you are expected to take steps to prevent that from reaching a court filing. The opinion explicitly states that lawyers "should not rely on GAI outputs without independent verification or review."

Confidentiality (Model Rule 1.6). Inputting client information into AI tools may constitute disclosure of confidential information to a third party. The opinion says boilerplate consent language in engagement letters is insufficient. Lawyers must secure informed consent before using client confidences in AI tools.

Communication (Model Rule 1.4). Lawyers may need to disclose their use of AI to clients, particularly when it affects fees, involves sharing confidential data, or when the client asks.

Candor to the Tribunal (Model Rules 3.1 and 3.3). Filing AI-generated content that contains fabricated citations violates the duty of candor. This is the rule most directly implicated in the sanctions cases above.

Supervisory Responsibilities (Model Rules 5.1 and 5.3). Partners and supervising lawyers must establish firm-wide policies on AI use and ensure that all lawyers and staff are trained on the risks.

Fees (Model Rule 1.5). Billing clients for time spent correcting AI errors, or charging full research rates for AI-assisted work, raises reasonableness concerns.

The bottom line from the ABA: you can use AI, but you own every word it generates. There is no "the AI did it" defense.

State-Level Rules Are Proliferating

Beyond ABA guidance, state bars and individual courts are rapidly adopting their own AI rules. By mid-2026, 35 state bar associations had issued guidance directing attorneys to verify AI output before relying on it (ComplianceHub, 2026).

In California, multiple federal judges and at least one state court judge now require disclosure and certification of accuracy when AI is used to prepare court documents.

Lawyers must identify the specific tool used, describe which portions of the filing were affected, and certify that a human lawyer reviewed the content.

Pennsylvania mandates explicit disclosure of AI use in all court submissions.

North Carolina issued Formal Ethics Opinion 2024-1, which requires lawyers to maintain competence in legal technology and thoroughly vet AI vendors. Oregon followed with a similar opinion in 2025.

The Charlotte Division of the Western District of North Carolina issued a standing order in June 2024 requiring certification that either no generative AI was used (with exceptions for established platforms like Westlaw and Lexis) or that every statement and citation was verified by a human.

The trend is clear: courts and bar associations are not waiting for a unified federal standard. They are implementing requirements jurisdiction by jurisdiction.

A Practical Verification Protocol

Here is a step-by-step checklist you can implement immediately. Print it out. Tape it next to your monitor. Make it part of your workflow before any AI-assisted filing.

Step 1: Never Use Raw ChatGPT for Case Research

General-purpose chatbots (ChatGPT, Claude, Gemini) are not legal research tools. They are language models.

Using them to find supporting case law is like asking a creative writing professor to do your Westlaw search. They will give you something that reads beautifully and cites nothing real.

If you are going to use AI for legal research, use a tool specifically designed for it, one with a retrieval layer connected to an actual case law database.

Step 2: Verify Every Citation Against a Primary Source

Every single case citation in an AI-generated output must be checked against Westlaw, Lexis, Google Scholar, PACER, or the court's own records. Not spot-checked. Every one.

Check three things:

• Does the case exist? Search the exact citation. If it returns nothing, it is fabricated.
• Does it say what the AI claims? Read the actual opinion. Confirm the holding matches what the AI attributed to it.
• Is it still good law? Run a Shepard's or KeyCite check. A real case that has been overruled is almost as bad as a fake one.

Step 3: Verify Statutory Citations Too

AI tools do not only fabricate cases. They also misquote statutes, cite repealed provisions, and invent subsections. Cross-check every statutory citation against the current code.

Step 4: Check Quotations Word-for-Word

If the AI puts quotation marks around language and attributes it to a court, verify the exact wording. AI-generated "quotations" are frequently paraphrases, or entirely invented text presented in quotation marks.

This is particularly dangerous because a fabricated quote attributed to a specific judge in a specific opinion is a misrepresentation to the tribunal.

Step 5: Document Your Verification

Keep a record of what you verified and how. If a judge questions your citations, you want to show a clear paper trail demonstrating due diligence. This is especially important in jurisdictions with AI disclosure requirements.

Step 6: Have a Second Set of Eyes

Before filing, have another lawyer or a trained legal research professional review the AI-assisted portions of your brief independently. This is the same quality control you would apply to work from a junior associate. AI output deserves at least the same scrutiny.

Step 7: Comply with Local AI Disclosure Rules

Check whether your jurisdiction requires AI disclosure. If it does, comply. If you are unsure, disclose anyway. No court has ever sanctioned a lawyer for being too transparent about their research methods.

What to Look for in a Legal AI Tool

Not all AI legal research tools carry the same risk. When evaluating a tool, ask these questions:

Does it search an actual case law database? A tool that generates answers from an LLM's training data alone is a hallucination machine. You need retrieval-augmented generation backed by a real, comprehensive case law corpus.

Ask the vendor: how many cases are in your database? How often is it updated? What courts does it cover?

Does it show its sources? Every claim in an AI-generated legal research output should link directly to the source case or statute. If the tool gives you a legal conclusion without showing you where it came from, you cannot verify it, and you should not trust it.

Does it provide the actual source text? Linking to a case is good. Showing you the exact paragraph from the judgment is better. The ability to see the source text alongside the AI's summary lets you quickly confirm accuracy without opening a separate database.

Does it flag confidence levels? Some legal AI tools provide confidence scores or verification indicators for each claim. This is a meaningful safety feature.

It tells you where the AI is highly confident (because it found a direct, on-point source) versus where it is less certain (because the retrieval was weaker).

Does it cross-check its own output? The most reliable approach is multi-layer verification. The system retrieves and summarizes, then actively checks each claim against the source material rather than trusting its own first draft.

Some tools run a multi-layer verification process inside the app (exact text matching, citation checking, semantic analysis, and cross-verification) with confidence labels (Verified, Partially Verified, Unverified) so you can see where to focus manual review.

Can you access the original court document? The gold standard is a tool that lets you click through to the actual court-copy PDF or official text of the judgment. If you can read the primary source within the same workflow, verification becomes fast instead of painful.

Does the vendor disclose hallucination rates? Any vendor that claims zero hallucinations is either lying or has not tested rigorously. Look for vendors that publish accuracy benchmarks and are transparent about limitations.

The verification habit that quietly fails most firms

Talk to litigators who run sanctioned cases and the same pattern surfaces: the lawyer who got hit was not the one ignoring verification. They were the one spot-checking.

Pull a brief, open the two highest-stakes citations, confirm they exist, sign off on the rest. That habit looks responsible and is exactly the failure surface in Mata and Wadsworth. The fabricated cites are not always the most prominent ones; they are scattered throughout the brief at the spots a tired associate trusted the formatting.

Spot-checking looks like verification and behaves like a gap. The fix is a workflow change. Check every cite, in order, against the source. Telling people to "check harder" does nothing.

The same applies to confidence scores. Some tools surface a per-claim "verified / partially verified" label. Treating that label as the verification step is exactly the trap: the model is grading its own homework.

The label is useful as triage (start with "partially verified" claims) but useless as a stop sign.

The Real Risk Is Not AI Itself

The sanctions cases above share a common feature: none of the lawyers were sanctioned for using AI. They were sanctioned for not verifying AI output before filing it with a court.

Judge Castel did not say lawyers cannot use ChatGPT. He sanctioned Schwartz and LoDuca because they submitted fabricated cases and then doubled down when challenged.

Judge Rankin did not ban AI in the Wyoming courtroom. He sanctioned the Morgan and Morgan lawyers because none of them read the cases they cited.

The ABA has been explicit about this. Formal Opinion 512 treats AI as a tool, like any other, that triggers existing professional obligations. The obligation to verify your research existed long before ChatGPT.

Lawyers who relied on a junior associate's memo without checking the citations were always taking a risk. AI just made it easier to generate large volumes of unchecked work product.

The practical implication: AI can make you faster and more thorough, but only if you build verification into your workflow. Without verification, AI makes you faster at producing sanctionable filings.

What to Do Next

If you are already using AI for legal research, or planning to start, here are concrete next steps:

1. Audit your current AI usage. What tools are your lawyers using? Are they using general-purpose chatbots for case research? If so, stop that immediately and switch to purpose-built legal AI tools with retrieval layers.

2. Establish a firm-wide AI policy. ABA Formal Opinion 512 specifically requires managing partners to create clear policies. At minimum, the policy should specify approved tools, require verification of all AI-generated citations, and mandate compliance with local disclosure rules.

3. Train your team. Every lawyer and paralegal who touches AI output needs to understand hallucination risks and verification procedures. This is not a one-time CLE. It should be ongoing as tools evolve.

4. Pick the right tools. Evaluate legal AI platforms based on the criteria above. Prioritize tools that show sources, provide confidence indicators, and connect to verified case law databases. The Stanford research shows that purpose-built legal RAG tools hallucinate at significantly lower rates than general-purpose models, but they still require verification.

5. Track your jurisdiction's rules. AI disclosure requirements are changing quickly. Assign someone in your firm to monitor state bar opinions and local court orders in every jurisdiction where you practice.

6. Document everything. Keep records of your AI usage, your verification steps, and your compliance with disclosure requirements. If a question ever arises, you want a clear record showing that you treated AI output with the same professional skepticism you would apply to any other research source.

The lawyers who get sanctioned are not the ones who use AI. They are the ones who trust it blindly.

Verification is the difference between a powerful research tool and a career-ending liability.

FAQ

Can a lawyer be sanctioned for using AI?

No lawyer in the cases above was sanctioned for using AI. They were sanctioned for filing fabricated citations they never verified. Rule 11 and ABA Formal Opinion 512 make the human signer responsible for every word, so AI use is fine as long as you check the output against a primary source.

What was the Mata v. Avianca case?

Mata v. Avianca, Inc. (1:22-cv-01461, S.D.N.Y., 2023) was the first high-profile AI hallucination sanction. Lawyers Steven Schwartz and Peter LoDuca filed a brief with six cases ChatGPT invented, then submitted the chatbot's false assurance that the cases were real. Judge Castel fined them $5,000 and found bad faith under Rule 11.

What is the largest AI hallucination sanction so far?

The largest known U.S. penalty is $110,000, imposed in December 2025 against San Diego attorney Stephen Brigandi and Portland attorney Tim Murphy in the District of Oregon. Their briefs contained 15 nonexistent cases and eight fabricated quotations (ABA Journal, 2026).

How much are AI hallucination sanctions?

They range from $1,000 to $110,000 in reported U.S. cases, plus non-monetary penalties like suspension and disbarment. U.S. courts imposed more than $145,000 in AI hallucination sanctions in the first quarter of 2026 alone (ComplianceHub, 2026). Amount tracks conduct: courts punish the cover-up harder than the initial error.

Have federal appeals courts sanctioned lawyers for AI hallucinations?

Yes. In 2026 the Fifth Circuit fined an attorney $2,500 for fabricated authority produced with vLex and CoCounsel, the Sixth Circuit hit two Tennessee lawyers with $15,000 each under Rule 38 in Whiting v. City of Athens, and the Ninth Circuit fined two California attorneys $2,500 each and suspended them from the court for six months in Lnu v. Blanche. AI hallucination sanctions are no longer confined to trial courts.

Can I be sanctioned if I only signed the brief but did not write it?

Yes. Federal Rule 11 attaches to the signature. In Wadsworth v. Walmart the two lawyers who signed but did not draft the motion were each fined $1,000, and in the Oregon case a local-counsel attorney who signed papers he had not written was held fully accountable. Signing certifies you made a reasonable inquiry into the law, whether or not you touched the keyboard.

How many lawyers have been caught citing fake AI cases?

Damien Charlotin's tracker counted 1,624 cases worldwide as of 18 June 2026, the most complete public count. The rate has climbed from a few a month in early 2025 to several a day, and a large share involve licensed attorneys rather than self-represented litigants.

Does ChatGPT make up legal cases?

Yes. General-purpose chatbots predict plausible text, not verified records. Stanford HAI measured hallucination rates of 58 to 88% on raw large language models. Even purpose-built legal AI tools hallucinated 17 to 33% of the time in Stanford's 2024 testing, so every citation still needs checking.

How do I verify AI legal citations before filing?

Check every citation against Westlaw, Lexis, Google Scholar, PACER, or the court's own records. Confirm the case exists, that it says what the AI claims, and that it is still good law. Then verify quotes word for word and document what you checked. The seven-step protocol above walks through it.

Are general-purpose chatbots ever safe for legal research?

They are useful for drafting and brainstorming, not for finding authority. If you need supporting case law, use a tool with a retrieval layer tied to a real case law database, and verify the output anyway.

Verify before you file

Vaquill AI is built so verification is fast instead of painful. It runs legal research over an actual case law and statute corpus, shows the source text beside every claim, and labels each citation Verified, Partially Verified, or Unverified so you know where to focus manual review. We do not claim zero hallucinations; no honest vendor does. We make the check quick enough that you actually do it on every cite. See how the citation verification works.

For related coverage, see How to Verify AI Legal Citations Before You File, the Charlotin hallucination tracker read like a risk manager, ABA Formal Opinion 512, the lawyer's guide, the Oregon $110,000 sanction breakdown, and legal AI that avoids hallucinating cases.

Arshita Anand

Co-Founder & CEO · Attorney

Arshita leads product and strategy at Vaquill, building the legal AI suite that solo, small-firm, and in-house US lawyers use to run a matter end to end.