AI Made It Up. Your Signature Makes It Your Problem

A lawyer asks an AI tool for cases supporting an argument.

The answer looks excellent.

It provides case names. Reporter citations. Quotations. Pinpoints. A convincing explanation of the legal rule.

There is only one problem:

Some of it never happened.

The cases may not exist. The quotation may not appear in the cited decision. The case may be real but stand for something completely different.

And once that information appears in a brief, client communication, memorandum, demand letter, website article, or other professional work product, saying “the AI gave it to me” does not make the problem disappear.

Courts have increasingly made that point explicit.

In June 2026, the U.S. Court of Appeals for the Ninth Circuit explained that lawyers were not being sanctioned simply because they used generative AI. The problem arose when attorneys signed and filed material containing nonexistent cases, fabricated quotations, and serious misrepresentations without adequate verification.

That distinction matters.

The lesson for law firms is not:

Never use AI.

It is:

Never treat AI-generated information as verified legal information simply because the answer sounds convincing.

For attorneys, paralegals, intake staff, legal assistants, marketers, and anyone else working around client matters, AI should be used deliberately, within defined boundaries, and with meaningful human review.

Because the consequences are no longer hypothetical.

How Big Is the AI Hallucination Problem in Law?

There is currently no official global registry recording every legal matter affected by an AI hallucination, so claims that there have been exactly a certain number of sanctioned lawyers should be treated cautiously.

However, the scale is substantial.

A widely cited database maintained by researcher Damien Charlotin currently identifies 1,870 legal decisions worldwide in which courts or tribunals address AI use or hallucinated material beyond a passing reference. The database itself cautions that it is a work in progress and includes some decisions where AI use was alleged but not conclusively established, so the figure should not be interpreted as 1,870 sanctioned lawyers or 1,870 monetary fines.

Separately, a June 2026 research paper examining legal citation hallucinations reported finding more than 1,000 filings containing fabricated citations, with the number increasing year over year.

The problem has also moved beyond obviously fictional cases.

A 2026 Fifth Circuit opinion warned that modern hallucinations increasingly appear as false quotations or inaccurate statements attributed to real cases, which can be harder to detect than a completely invented decision.

That may be the more dangerous version.

A fake case can often be discovered by searching for it.

A real case containing a fake quotation can survive a much more casual review.

Real Lawyers. Real AI Hallucinations. Real Consequences.

1. Mata v. Avianca: $5,000 After Seven Nonexistent Cases

One of the cases that first brought widespread attention to AI hallucinations in legal practice was Mata v. Avianca.

A later Fifth Circuit opinion recounting the incident noted that the AI-generated brief cited seven nonexistent cases, including a fictional decision called Varghese v. China Southern Airlines Co. The lawyers in Mata ultimately received a $5,000 sanction.

That case became famous.

It did not stop the problem.

2. Fletcher v. Experian: $2,500 for Unverified AI Drafting

In February 2026, the Fifth Circuit sanctioned attorney Heather Hersh in Fletcher v. Experian Information Solutions, Inc.

The court determined that AI had been used to draft a substantial portion, potentially all, of a reply brief and that counsel failed to verify the generated content. It also found that counsel had not been forthcoming when responding to the court's inquiry.

The result:

$2,500 in sanctions.

The court made another important observation: as AI becomes more sophisticated, inaccurate output can become more difficult to recognize because it increasingly resembles legitimate legal writing.

3. Lnu v. Blanche: Fines, Six-Month Suspensions, Bar Notification

The consequences became significantly more serious in the Ninth Circuit's June 2026 decision in Lnu v. Blanche.

The court found briefs containing:

• Nonexistent cases
• Misattributed quotations
• Gross misrepresentations of actual cases
• Repeated failures of candor when the errors were questioned

The court stressed that it was not punishing the lawyers simply for using generative AI. It was disciplining them for signing and filing inaccurate work and for their subsequent conduct.

The consequences were extensive.

Each attorney was:

• Fined $2,500
• Suspended from practicing before the Ninth Circuit for six months
• Ordered to provide the disciplinary order to clients, opposing counsel, judges in pending matters, and every lawyer in their firm
• Required for two years to disclose generative AI use in future filings and certify personal review of citations and quotations

The court also directed that the order be sent to the California State Bar and other applicable licensing authorities.

A bad AI citation had become much more than a citation problem.

It became a professional-discipline problem.

4. Lexos Media IP v. Overstock: Multiple Lawyers Sanctioned

AI mistakes do not necessarily stop with the person who typed the prompt.

In Lexos Media IP, LLC v. Overstock.com, Inc., a federal court in Kansas addressed filings containing fabricated and misrepresented legal authorities created through generative AI and not properly checked before filing.

One attorney admitted using ChatGPT to assist with research and drafting without confirming the authorities it produced. But the court also focused on the responsibilities of the other lawyers whose names appeared on the filings.

The sanctions included:

• $5,000 against the attorney who used ChatGPT, plus revocation of his pro hac vice admission and a requirement to self-report to disciplinary authorities
• $3,000 against another attorney
• $3,000 against another attorney
• $1,000 against local counsel
• Public admonishment of all five attorneys involved

The court also required internal procedures designed to improve verification of legal authorities.

That is $12,000 in monetary fines across four attorneys, before considering the professional, administrative, and reputational consequences.

The same order reviewed prior AI-related sanction cases and noted monetary consequences ranging from $500 to $31,000 in cases the court considered.

This is why a law firm's AI policy cannot apply only to the person using ChatGPT.

Supervision matters too.

5. A 2026 Connecticut Case: $3,500 and a Grievance Referral

In August 2026, a federal judge in Connecticut sanctioned attorney Hilary Miller after identifying nonexistent quotations, erroneous citations, and misrepresentations in filings that had incorporated AI-generated material.

The lawyer had attempted to verify the cases through Lexis, but the court found that the review did not adequately verify the precise quotations and pin cites.

The result was a $3,500 monetary sanction and referral to the court's Grievance Committee. The court also terminated one pending motion, allowed corrected briefing, and imposed additional verification requirements for future filings.

This case contains an especially important lesson:

Checking whether a case exists is not enough.

You must verify:

• The quotation
• The pinpoint citation
• The proposition attributed to the case
• The actual holding
• The jurisdiction
• Whether the case remains good law

AI verification cannot be superficial.

Canada Is Dealing With the Same Problem

This is not exclusively an American issue.

Canadian courts and professional organizations have also warned legal professionals about AI-generated authorities.

In Zhang v. Chen, a British Columbia lawyer relied on two nonexistent cases generated through ChatGPT. The court addressed personal cost consequences arising from the use of fictitious authorities. The case has since become one of Canada's best-known examples of the professional risks of unverified generative AI research.

In Ontario's R. v. Chand, the court identified fictitious or erroneous authorities in defence submissions and ordered new submissions with citations checked and hyperlinked to authoritative sources. The court specifically directed that generative AI, including commercial legal software using generative AI, not be used for the new legal research in that matter.

Alberta's courts have also formally instructed practitioners and litigants to use authoritative sources for legal authorities and require meaningful “human in the loop” verification of AI-generated submissions.

The Canadian Bar Association likewise cautions that generative AI cannot independently validate the accuracy of its output and may confidently provide information that does not align with reality.

Why Does AI Hallucinate So Convincingly?

The problem is partly psychological.

AI-generated legal writing often looks right.

The case name sounds plausible.

The quote sounds judicial.

The citation follows the expected format.

The explanation uses terminology lawyers recognize.

That fluency can create misplaced confidence.

The ABA's Formal Opinion 512 warns that generative AI systems can produce fluent text while lacking the ability to understand the meaning and context of what they generate. The result can be unreliable information or plausible responses with no basis in fact.

That means an AI chatbot should not be thought of as a junior associate who “knows the law.”

A better analogy is:

AI is an extraordinarily fast drafting engine that sometimes produces confident fiction.

The professional using it still has to determine which is which.

Even Legal AI Is Not Automatically Hallucination-Free

Using a legal-specific AI product can reduce some risks compared with blindly asking a general-purpose chatbot for case law.

It does not eliminate the verification requirement.

A Stanford-led preregistered study published before many of today's newest models evaluated several leading AI legal research products and found hallucination rates between 17% and 33% in the systems tested. The ABA subsequently cited that research in Formal Opinion 512.

Those figures should not be treated as current hallucination rates for every legal AI platform in 2026. Models and products have changed substantially since that evaluation.

The important finding is simpler:

A product being designed for legal research does not eliminate the need for professional verification.

The Risk Goes Beyond Fake Citations

When law firms discuss AI mistakes, fake cases receive most of the attention.

But casual AI use creates several separate risks.

1. Incorrect Legal Advice

An AI system may:

• Apply the wrong state's law
• Blend U.S. and Canadian legal principles
• Use outdated legislation
• Misstate a limitation period
• Invent an exception
• Oversimplify a legal test
• Apply the correct rule to the wrong jurisdiction

If that information reaches a client without qualified review, the problem is no longer merely technological.

It is professional work product.

2. Confidentiality and Client Data

Another common mistake is casually copying a client's:

• Medical records
• Employment records
• Financial documents
• Discovery responses
• Emails
• Contracts
• Names
• Addresses
• Strategy
• Settlement information

into an AI tool without first understanding how that system stores, processes, or uses the information.

ABA Formal Opinion 512 specifically identifies confidentiality and unauthorized disclosure as risks lawyers must evaluate when using generative AI.

Canadian guidance similarly emphasizes the need to understand how client information is handled before placing confidential data into third-party AI systems.

The question should therefore never be simply:

“Can ChatGPT summarize this?”

The first question is:

“Are we permitted to put this information into this particular system?”

3. Supervisory Responsibility

A firm partner may never personally open an AI chatbot and still have an AI problem.

A paralegal might use one.

A junior associate might use one.

An intake coordinator might use one.

A marketing contractor might use one.

An external co-counsel might use one.

The District of Kansas decision in Lexos Media specifically emphasized managerial and supervisory responsibilities involving lawyers and nonlawyers and pointed to ABA guidance requiring firms to establish policies and training around generative AI.

A policy that simply says “Do not use ChatGPT” is rarely enough if nobody is trained, nobody verifies compliance, and employees quietly continue using personal AI accounts.

So Should Law Firms Stop Using AI?

No.

That would miss the point.

The Ninth Circuit stated clearly in 2026 that there is nothing inherently improper about appropriately using AI in legal practice.

The better approach is to separate productive AI assistance from unverified AI authority.

A useful internal principle is:

Use AI as a Drafting Layer, Not a Source Layer

AI is generally safer when it is transforming information you already possess and can verify.

It becomes much riskier when you ask it to supply authoritative information you have not independently confirmed.

Safer Ways Legal Professionals Can Use AI

Depending on the tool, firm policy, client obligations, and applicable rules, appropriate uses may include:

Brainstorming

Ask AI to generate:

• Possible issues to investigate
• Interview questions
• Alternative headings
• Arguments to research
• Checklist ideas
• Questions that may be missing from an intake form

Then conduct actual legal research independently.

Drafting From Verified Material

Provide approved source material and ask AI to:

• Improve organization
• Convert notes into a first draft
• Simplify language
• Create an outline
• Compare two versions
• Turn a meeting summary into action items

The key difference is that the source material came from you, rather than the model inventing the underlying facts or authorities.

Summarizing Documents

AI can help create preliminary summaries of lengthy materials, provided:

• The tool is approved for the information involved
• Confidentiality requirements have been considered
• Important conclusions are checked against the original document
• The summary is not treated as a substitute for reviewing critical evidence

Administrative Workflows

AI may assist with lower-risk operational tasks such as:

• Categorizing inquiries
• Drafting internal templates
• Preparing meeting agendas
• Organizing non-sensitive information
• Generating process documentation
• Converting structured information into standard internal formats

Human review should still remain appropriate to the task.

Marketing and Educational Content

AI can accelerate:

• Blog outlines
• Social media concepts
• FAQ structure
• Newsletter drafts
• Topic research
• Content repurposing

But legal statements, statistics, regulatory claims, case references, and jurisdiction-specific information should be independently sourced and verified before publication.

A hallucinated case on your website may not result in a Rule 11 sanction.

It can still damage the firm's credibility.

High-Risk AI Uses That Deserve Much More Control

Law firms should be especially cautious about using general-purpose AI to:

• Find cases and citations that will be relied on without independent research
• Generate quotations from judicial opinions
• Determine filing deadlines
• Provide final legal advice
• Determine whether a conflict exists
• Assess whether representation should be accepted
• Analyze highly confidential material in an unapproved system
• Calculate damages without verifying inputs and methodology
• Interpret statutes without checking the current text
• Produce a final court filing without line-by-line review
• Communicate directly with clients without appropriate supervision
• Make decisions requiring professional legal judgment

AI can assist with many of these workflows.

It should not silently become the final decision-maker.

The Five-Question AI Check for Law Firms

Before using AI-generated material in professional work, ask:

1. Where did this information come from?

Can you trace the statement to an authoritative source?

2. Have I personally checked the source?

Do not verify a citation by asking the same AI whether the citation is real.

Open the actual case, statute, regulation, contract, record, or authoritative database.

3. Is this tool approved for the information I am entering?

Consider confidentiality, privacy, data retention, vendor access, and client obligations.

4. Is AI performing an assistive task or making a professional judgment?

The closer the task moves toward legal advice, case strategy, conflicts, deadlines, or final representation decisions, the more human oversight it requires.

5. Would I be comfortable signing my name underneath this output?

Because eventually, that is the standard that matters.

Every Law Firm Needs an AI Policy Now

An effective AI policy does not need to be 40 pages long.

It does need to answer practical questions.

At minimum, staff should know:

• Which AI tools are approved
• Which tools are prohibited
• What client information may be entered
• What information must never be entered
• Whether client consent is required
• Which tasks AI may assist with
• Which tasks require attorney review
• How legal citations must be verified
• How AI-generated quotations must be checked
• Whether AI use must be disclosed
• Who is responsible for supervising staff
• What to do when an AI error is discovered
• How outside vendors and contractors may use AI
• Whether personal AI accounts may be used for firm work

ABA Formal Opinion 512 emphasizes competence, confidentiality, communication, candor, supervision, and reasonable fees when lawyers use generative AI.

The safest firms will not merely purchase AI.

They will define how it is allowed to enter the workflow.

What Should You Do If You Discover an AI Hallucination?

Do not try to hide it.

Recent court decisions increasingly distinguish between the original mistake and the professional's conduct after discovering it.

Courts have treated prompt disclosure, responsibility, correction, and cooperation as mitigating factors, while evasiveness and lack of candor have contributed to harsher consequences.

If potentially inaccurate AI-generated information has entered legal work:

1. Stop relying on it.
2. Verify the underlying source.
3. Identify everywhere the incorrect information was used.
4. Escalate the issue internally.
5. Determine whether correction, disclosure, client communication, or other action is required.
6. Document what happened.
7. Fix the workflow that allowed the error to pass through.

The exact professional obligations depend on the jurisdiction and circumstances and should be evaluated by qualified counsel.

AI Is Powerful. Casual AI Use Is the Problem.

Generative AI can save legal professionals substantial time.

It can help structure information, accelerate drafting, automate repetitive work, improve internal processes, and make firms more efficient.

But efficiency only matters when the underlying work remains reliable.

A fabricated citation generated in three seconds can require hours of corrective work.

It can produce:

• Monetary sanctions
• Adverse cost awards
• Disciplinary referrals
• Suspension from practice before a court
• Withdrawal or termination of filings
• Additional work for clients and opposing counsel
• Reputational damage
• Confidentiality problems
• Potential professional liability

And perhaps most importantly, it can hurt the person who hired the firm expecting professional judgment.

Your client did not hire your AI tool.

They hired your judgment.

AI can help prepare the work.

The legal professional must still own the result.

Not Sure How AI Should Fit Into Your Firm?

The answer is not necessarily buying another AI subscription.

For many firms, the first step is deciding:

• What should be automated
• What should remain human
• Which tools are appropriate
• Where verification must occur
• How staff should use AI
• How confidential information should be handled
• How AI fits into the firm's existing workflow

SyncReach works with law firms on practical digital systems, automation, integrations, and consulting designed around how the firm actually operates.

If your team is already using AI but you are not certain whether everyone is using it safely, that is exactly the kind of question worth addressing before an error reaches a client, a court, or opposing counsel.

Book a free consultation with SyncReach to discuss how AI and automation can fit into your firm's workflow without removing the human judgment legal work depends on.

Frequently Asked Questions

Can lawyers use ChatGPT or other generative AI tools?

Yes. The use of generative AI is not inherently prohibited. The professional obligations that apply to a lawyer's work continue to apply when AI assists with that work. Courts and professional bodies emphasize verification, competence, confidentiality, candor, and human judgment.

Can lawyers be fined for AI hallucinations?

Yes. Courts have imposed monetary sanctions when lawyers submitted inaccurate or fabricated legal authorities associated with unverified AI use. Examples discussed in this article include $2,500, $3,500, $5,000, and larger combined sanctions involving multiple attorneys. One 2026 federal court order reviewing prior cases noted monetary penalties ranging from $500 to $31,000 among the decisions it considered.

How many legal cases have involved AI hallucinations?

There is no authoritative global total. Researcher Damien Charlotin's widely cited tracker currently lists 1,870 legal decisions involving AI use or hallucinated material, but the database expressly notes that it is a work in progress and includes some matters where AI use was alleged but not confirmed. It should therefore not be interpreted as 1,870 sanctioned lawyers.

Can legal-specific AI hallucinate?

Yes. Legal-specific products can reduce certain risks but are not automatically error-free. A Stanford-led study of several legal AI research products found hallucination rates between 17% and 33% in the systems tested at that time. Because AI products continue to evolve, those figures should not be used as current rates for all 2026 tools.

What is the safest way for a law firm to use AI?

A useful general rule is to use AI to assist with drafting, organization, summarization, brainstorming, and workflow tasks while independently verifying authoritative information and maintaining human control over professional judgment.

Firms should also have clear policies regarding approved tools, confidentiality, verification, staff supervision, and client information.

This article is provided for general educational and marketing purposes and does not constitute legal, ethics, cybersecurity, privacy, or professional-responsibility advice. Rules governing AI use differ by jurisdiction, court, practice area, and factual circumstances. Law firms and legal professionals should review applicable court rules, professional-conduct obligations, privacy requirements, and relevant guidance before implementing AI tools.

← Back to all articles