When the Case Exists but the Legal Analysis Is Wrong
Why verifying citations is only the beginning, and how attorneys can establish reliable, defensible processes for AI-assisted legal work.
Analysis by
Founder, Lex Wire Journal • Technology, Governance & Sovereignty Strategist
A lawyer receives an AI-generated research memorandum addressing a dispositive legal issue. The document is organized, persuasive, and supported by citations to published judicial opinions.
The lawyer checks the citations. Every cited case exists. The quotations are accurate. The memorandum appears ready for incorporation into a motion.
There is only one problem.
One cited case discusses a party’s rejected argument, but the memorandum presents that argument as the court’s holding. Another decision comes from a jurisdiction whose law does not control the dispute. A third contains a material factual distinction that the AI-generated analysis never addresses.
The citations are authentic. The legal conclusion is unreliable.
Has the lawyer verified the work?
In one limited sense, yes. The lawyer has confirmed the existence of the authorities. But the lawyer has not established that those authorities support the propositions advanced or that the reasoning properly applies to the client’s situation.
That distinction represents a central challenge of artificial intelligence in legal practice.
Much of the profession’s discussion has focused on hallucinations: invented cases, fabricated quotations, and nonexistent legal authorities.
Those errors are serious, but they represent only part of the problem.
An AI system can identify a real case while mischaracterizing its holding. It can accurately summarize one portion of an opinion while overlooking a qualification that changes the result. It can present incomplete research as a comprehensive legal answer.
The resulting analysis may appear professionally credible because its individual components are real.
That makes the error less obvious, not less consequential.
In my continuing legal education program, Ethics in the Age of AI, I emphasize that attorneys must independently verify AI-assisted legal analysis rather than accepting plausible-looking output.
“Check every case citation, not just that the case exists, but that it actually says what the AI claims it says.”
Jeff Howell, Esq. | Ethics in the Age of AI, Original CLE Instructional Materials
The need for substantive verification received significant judicial attention in 2026.
In LNU v. Blanche , the Ninth Circuit examined not only nonexistent authorities but also inaccurate descriptions of genuine judicial decisions.
The court recognized that inaccuracies involving real authorities may be more difficult to detect than outright fabrications because identifying the problems requires meaningful examination of the sources.
The lesson extends well beyond appellate briefing.
Lawyers using AI to analyze contracts, evaluate litigation risk, conduct due diligence, or prepare client advice face the same fundamental question:
What makes AI-assisted legal work sufficiently reliable for a lawyer to stand behind it?
The answer requires more than software accuracy.
It requires a professional process that connects authentic evidence, sound legal reasoning, and accountable attorney review.
The Bottom Line
Lawyers must do more than confirm that AI-generated citations exist. They must evaluate whether sources actually support the propositions asserted, whether the analysis applies the correct legal standards, and whether material facts or adverse authorities have been omitted.
A defensible AI-assisted workflow requires three related elements: Source Integrity, Substantive Verification, and Process Accountability.
These are proposed components of a practical review framework, not a new ABA ethics rule. The appropriate review process depends on the work performed, applicable legal requirements, and the risks involved.
Beyond Hallucinations: The Real Problem Is Unverified Legal Reasoning
The term hallucination is often associated with AI systems generating information that does not exist.
In legal practice, the most recognizable example is a fabricated judicial opinion accompanied by a convincing citation.
But legal unreliability takes several different forms.
Some errors concern the authenticity of the underlying source. Others concern the meaning or legal relevance of a genuine source.
Still others involve omissions, unsupported assumptions, or the incorrect application of legal standards to facts.
These distinctions matter because the verification methods necessary to detect each error differ.
| Error Type | Illustrative Failure | Verification Requirement |
|---|---|---|
| Fabricated Authority | A nonexistent case or statute | Confirm the primary source |
| Mischaracterized Holding | A genuine opinion cited for a proposition it does not establish | Read the decision and evaluate its reasoning |
| Incorrect Quotation | Real authority accompanied by inaccurate quoted language | Compare the quotation to the original |
| Outdated Authority | Reliance on law that has been reversed, limited, or superseded | Validate subsequent treatment and current law |
| Omitted Authority | Failure to identify material controlling or contrary law | Conduct independent and appropriately comprehensive research |
| Faulty Application | A sound general rule applied to materially different facts | Evaluate factual distinctions and legal elements |
| Unsupported Fact | A factual claim presented without an adequate evidentiary basis | Check records, testimony, exhibits, or other appropriate evidence |
The categories overlap, but they illustrate the limitations of a simple citation-checking policy.
A case verification tool might establish that an opinion exists without determining whether the opinion addresses the legal issue for which it is cited.
A legal database may provide accurate source material without guaranteeing that an AI-generated synthesis of those materials is correct.
And an AI system that produces coherent reasoning may omit important evidence or controlling law.
The professional obligation is not to eliminate every possibility of human or technological error.
It is to perform the inquiry reasonably required by the task, the governing rules, and the circumstances.
From Mata to 2026: How Courts Are Addressing AI Verification Failures
Judicial decisions involving AI-assisted filings demonstrate that courts are applying existing professional and procedural obligations to new methods of producing legal work.
Three decisions are particularly useful for understanding the progression from fabricated authorities to broader accountability for inaccurate legal analysis.
Mata v. Avianca: Fabricated Authorities and Failure to Investigate
In Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), attorneys submitted nonexistent judicial opinions generated through ChatGPT.
The fabricated materials included false citations and purported judicial quotations.
After opposing counsel and the court questioned the authorities, the attorneys continued to represent the invented decisions as authentic.
The court imposed sanctions under Rule 11 or, alternatively, its inherent authority.
Its analysis emphasized that attorneys retain responsibility for checking the accuracy of their filings, regardless of the research technology used.
Mata demonstrates two separate failures.
The first occurred before filing: the attorneys failed to verify the authorities used to support their legal arguments.
The second concerned their response after the court called the questionable authorities to their attention.
This distinction matters because a lawyer’s responsibility does not necessarily end when a document is filed.
LNU v. Blanche: When Real Cases Support False Conclusions
In LNU v. Blanche, No. 24-4790 (9th Cir. June 3, 2026), the Ninth Circuit issued a published disciplinary order addressing briefs containing nonexistent cases, misattributed quotations, and serious misrepresentations of actual authorities.
The court explained that AI-related legal errors extend beyond invented citations.
A system may cite a genuine case while incorrectly describing the holding or treating a rejected party argument as the court’s conclusion.
It may also fail to distinguish between binding and persuasive authority.
The court emphasized that these more subtle inaccuracies can be difficult to detect without close reading of the underlying opinions.
The court imposed professional discipline on two attorneys, including monetary sanctions and suspension from practice before the Ninth Circuit.
The order also addressed the attorneys’ lack of candor concerning the source of the errors.
LNU is particularly important because it demonstrates why checking that a citation exists cannot be treated as complete legal verification.
Source authenticity is only one condition of reliable analysis.
A lawyer must also determine whether the cited authority supports the proposition, whether its legal status has been accurately described, and whether it properly applies to the dispute.
The court’s reasoning also reinforces a broader principle: professional obligations do not depend solely on whether a mistake originated with AI, a human assistant, or the attorney.
The origin of an error can matter to particular questions of candor, discipline, and remediation.
But it does not relieve lawyers of responsibility for the filings they submit.
LeDoux v. Outliers: Verification as an Express Court Requirement
In LeDoux v. Outliers, Inc., No. 3:24-cv-05808-TMC (W.D. Wash. July 24, 2026), the court addressed filings containing multiple inaccurate legal and factual citations generated with AI.
Counsel acknowledged using ChatGPT and Claude and failing to verify the generated citations against original sources before filing.
The errors also extended into citation material supplied for expert reports.
The court imposed a $3,000 personal sanction under Rule 11(b).
It additionally required counsel to include a certification in subsequent briefing in that case confirming that citations had been verified and accurately supported the propositions asserted.
The order also required counsel to provide the sanctions order to her client.
LeDoux demonstrates how verification failures may produce specific court-imposed obligations beyond the generally applicable professional rules.
The certification requirement arose from the circumstances of that case.
It should not be presented as a universal federal rule requiring identical certifications in every AI-assisted filing.
Together, Mata, LNU, and LeDoux demonstrate that the legal problem is not simply the use of AI.
It is the failure to perform appropriate independent verification and respond responsibly when errors are identified.
The Professional Rules Require More Than Citation Checking
The ethical framework governing AI-assisted legal work did not begin with generative artificial intelligence.
Lawyers have long been responsible for their research, factual representations, legal arguments, and supervision of professional work.
AI introduces new capabilities and failure modes, but existing rules remain central.
Rule 1.1: Competent Representation
ABA Model Rule 1.1 requires competent representation, including appropriate legal knowledge, skill, thoroughness, and preparation.
Competence in AI-assisted practice includes understanding the relevant tool’s capabilities and limitations, but it also includes examining the legal work produced.
An attorney who understands that AI can hallucinate has not necessarily satisfied the duty of competence.
The lawyer must translate that awareness into appropriate professional conduct.
Rule 3.3: Candor Toward the Tribunal
ABA Model Rule 3.3 addresses duties involving false statements, controlling adverse authority, and corrective action in specified circumstances.
Its requirements must be applied according to the rule’s particular knowledge elements and other conditions.
A mistaken citation, a knowing misrepresentation, and a failure to correct a statement after learning that it is false do not necessarily raise identical issues.
Rules 5.1 and 5.3: Supervision and Firm Responsibility
Rule 5.1 addresses the responsibilities of partners, managers, and supervisory lawyers.
Rule 5.3 concerns nonlawyer assistance.
Their application is particularly important when attorneys or staff use AI in workflows requiring professional review.
AI systems are not human associates, but their integration into supervised legal workflows requires lawyers to address the risks of relying on technology-generated work.
Federal Rule of Civil Procedure 11
Federal Rule of Civil Procedure 11 provides another essential source of accountability.
Presenting a pleading, written motion, or other covered paper constitutes specified certifications concerning its factual and legal basis after an inquiry reasonable under the circumstances.
The standard is not an absolute warranty that a legal argument will prevail.
It is a requirement of reasonable inquiry and appropriate factual and legal support.
These obligations apply regardless of whether a lawyer drafted the document manually, delegated portions to another professional, or used AI assistance.
ABA Formal Opinion 512 reinforces the need for lawyers to review generative AI outputs for accuracy and completeness in light of their professional duties.
Source Verification Versus Substantive Verification
A useful way to understand the professional problem is to separate two forms of verification.
Source verification concerns the authenticity and accurate representation of underlying material.
Substantive verification concerns whether the material supports the legal analysis and whether that analysis is appropriate to the matter.
Both may be necessary. Neither should be assumed to establish the other.
What Source Verification Establishes
For legal research, source verification includes confirming the existence of cases, checking quotations, confirming statutory text, and reviewing the current legal status of cited authorities.
For factual work, it may include checking contracts, transaction records, exhibits, testimony, and other source materials.
A properly verified citation should lead to the actual authority being relied upon.
What Substantive Verification Requires
Substantive verification requires the attorney to engage with the legal argument itself.
The lawyer should ask whether the authority actually supports the claimed proposition.
Is the proposition a court’s holding, a discussion of dicta, a party’s argument, or a quotation from another source?
Does the authority govern the jurisdiction? Has subsequent law modified its significance?
Are there material facts distinguishing the case from the client’s matter?
Has relevant contrary authority been omitted?
Does the argument overlook a statutory exception, procedural requirement, or unresolved question?
These questions cannot be answered merely by validating citation formatting or source existence.
“Use the AI research as a starting point, not an ending point.”
Jeff Howell, Esq. | Ethics in the Age of AI, Original CLE Instructional Materials
In my CLE, I discuss an approach called the AI Sandwich Method.
The lawyer may begin with AI-assisted brainstorming and identification of possible issues.
The attorney then conducts independent research through appropriate primary authorities and established legal research resources.
AI may subsequently assist in organizing or editing material whose underlying analysis has been verified.
The method is useful because it separates idea generation from authoritative research.
Its effectiveness, however, depends on the independence and quality of the verification stage.
Returning to AI to organize a verified analysis does not eliminate the need to review the final document for newly introduced errors.
The Defensible AI Work Product Framework
The professional challenge can be organized through three distinct dimensions of accountability.
I propose calling this the Defensible AI Work Product Framework .
The framework is intended to help attorneys evaluate whether AI-assisted legal work has received an appropriate level of professional review.
It is not an independently binding ethics standard or a judicially adopted verification test.
Establish that the underlying authorities, quotations, factual claims, and supporting evidence are authentic, current, and accurately represented.
Independently evaluate the legal reasoning, relevant facts, competing interpretations, material omissions, and application of governing law.
Establish an appropriate review workflow, assign professional responsibility, address exceptions, and maintain proportionate records when warranted.
Why Source Integrity Is Necessary
A lawyer cannot reasonably rely on an authority that has not been adequately identified or authenticated.
This is the lesson most directly associated with fabricated citations.
But source integrity is only the foundation, not the conclusion of professional verification.
Why Substantive Verification Is the Central Professional Task
Once a source is verified, the lawyer must determine whether it establishes what the AI claims.
That requires legal analysis, not merely mechanical checking.
The attorney must connect the relevant rule, authority, and facts through independent reasoning.
In many matters, this is precisely where the lawyer contributes the most professional value.
Why Process Accountability Matters
A reliable process clarifies who is responsible for reviewing work and when additional professional approval is required.
It can also help a firm identify recurring error patterns, training gaps, or weaknesses in its AI workflows.
However, keeping records does not automatically make an inadequate review reasonable or defensible.
Nor does a correct final result establish that the underlying inquiry necessarily satisfied professional obligations.
The three dimensions must operate together.
How Should Lawyers Verify AI-Generated Legal Research?
The level of verification should reflect the nature of the task, the importance of the legal conclusion, and the applicable professional requirements.
My CLE recommends tiered verification rather than treating every AI-assisted activity as presenting identical risk.
The following categories illustrate one possible internal firm approach.
| Risk Category | Illustrative Work | Suggested Review |
|---|---|---|
| Lower | General brainstorming using public information | Verify material claims before substantive reliance |
| Moderate | Internal research summary or preliminary contract review | Verify material sources, facts, and reasoning |
| High | Client-specific legal opinion or consequential strategy analysis | Independent substantive review and appropriate attorney approval |
| Critical | Court filings, material privilege determinations, evidence submitted to tribunals | Comprehensive task-appropriate verification under governing duties and court requirements |
These categories are a recommended management tool, not classifications mandated by the ABA Model Rules.
A firm should not use a risk label to justify submitting unverified authorities to a tribunal.
Even routine work may require thorough verification when it becomes part of a consequential legal filing.
Conversely, it may be unnecessary to apply the same elaborate review procedure to preliminary brainstorming that contains no client information and will not be relied upon.
The objective is to match the verification process to the actual professional risk.
Verification Is Not Agreement: The Value of Independent Challenge
One of the more difficult risks of AI-assisted work is that the output often appears coherent, complete, and persuasive.
A polished document can encourage reviewers to search for confirmation rather than examine what may be wrong.
The problem is not unique to AI.
Lawyers can become anchored to initial conclusions supplied by colleagues, clients, research summaries, or their own assumptions.
AI makes the problem especially relevant because it can rapidly generate detailed legal explanations.
In my CLE, I recommend challenge protocols that require reviewers to actively search for weaknesses.
“Create verification workflows that don’t rely on the same person who generated the AI content.”
Jeff Howell, Esq. | Ethics in the Age of AI, Original CLE Instructional Materials
A second reviewer can help identify assumptions that the original drafter may have overlooked.
But a separate reviewer is not always required, and the adequacy of review depends on the task.
The more important underlying principle is independence of professional analysis.
Lawyers can operationalize that principle by asking whether a different legal interpretation is available, whether contrary authorities exist, and whether a material factual change would alter the conclusion.
For complex matters, it may be useful to formulate an independent preliminary position before examining the AI recommendation.
That can reduce the risk that the first plausible AI-generated answer becomes the unquestioned foundation of the analysis.
Asking another AI system to check the first may provide useful questions or alternatives.
It does not, by itself, establish independent source verification.
Multiple AI systems can repeat the same incorrect proposition or rely on similar incomplete information.
The lawyer must still evaluate the underlying authorities and evidence.
Process Accountability: What Should Lawyers Document?
A defensible workflow benefits from clear responsibility and appropriately maintained records of review.
But documentation presents its own professional considerations.
A firm that retains every AI interaction without considering confidentiality, retention obligations, and discoverability may create risks unrelated to the original legal task.
The correct objective is not necessarily to preserve every prompt, response, and intermediate draft.
It is to establish records appropriate to the work, the applicable rules, and the firm’s information-governance policies.
Depending on the matter, relevant documentation may include the following.
Suggested AI Review Record
- Task: What AI-assisted activity was performed and why?
- Workflow: What system or approved process was used, where material?
- Reviewer: Which attorney or designated professional evaluated the work?
- Sources: Which material authorities, factual assertions, and supporting evidence were verified?
- Exceptions: What significant inaccuracies, omissions, or unresolved issues were identified?
- Approval: Who made the final professional decision, and was escalation required?
This is a proposed documentation model, not a universal regulatory requirement.
Some matters will require more extensive records. Others may require substantially less.
Where the firm retains sensitive prompts, client documents, or AI outputs, its retention practices should be evaluated in light of confidentiality and privilege risks.
Those concerns connect directly to AI and the Attorney-Client Privilege: Who Can Access Your Client’s Confidential Information? .
The central distinction is between keeping evidence of meaningful review and retaining information simply because a system makes it easy to collect.
When AI Verification Fails: Correction, Escalation, and Candor
A verification process must account for the possibility that an error will be discovered after work has already been circulated, relied upon, or filed.
The appropriate response depends on the nature of the error, its significance, and the applicable legal obligations.
A minor internal drafting mistake does not necessarily create the same duties as a false statement submitted to a tribunal.
Mata demonstrates how a failure to meaningfully investigate challenged authorities can compound the original problem.
LNU v. Blanche likewise emphasizes the importance of candor when AI-related inaccuracies are discovered.
In appropriate circumstances, the lawyer may need to investigate the scope of an identified defect, consult governing rules, notify relevant persons, correct a filing, or take other remedial action.
The precise requirements must be determined under applicable law, court orders, and professional rules.
Determine whether the issue concerns a fabricated source, inaccurate representation, faulty reasoning, or unsupported factual claim.
Determine whether the same defect affects other work, related arguments, or documents already distributed.
Evaluate professional obligations, procedural rules, court requirements, and necessary consultation.
Take required corrective steps promptly, including appropriate notification or filing.
Identify why the error escaped review and adjust training, verification, or supervision as appropriate.
The sequence is intended as a practical escalation framework.
It does not replace the specific corrective duties imposed by Rule 3.3, court orders, or other applicable law.
Three Practical Examples of Defensible AI-Assisted Legal Work
The following scenarios are hypothetical. They illustrate how the verification framework can be adapted to different forms of legal work.
The Authentic but Misleading Case
An attorney uses AI to prepare a memorandum concerning a statutory interpretation issue.
The AI cites an actual appellate decision. The citation is correct, but the memorandum treats a discussion of a rejected argument as the court’s holding.
Source verification establishes that the opinion exists.
Substantive verification requires the attorney to read the relevant portions of the decision and evaluate the court’s actual reasoning.
Appropriate process accountability identifies who performed the review and approved the conclusion.
The lesson is that citation accuracy is necessary but insufficient where the legal proposition itself has been misstated.
The AI Contract Review
A lawyer uses AI to review a commercial services agreement.
The system flags an unusually broad indemnification provision and summarizes the limitation-of-liability clause.
It fails, however, to recognize an interaction between the indemnification obligations and insurance coverage requirements.
The attorney cannot assume that the issues identified by AI represent every material contractual risk.
Appropriate verification may require reviewing relationships among provisions, considering the client’s business objectives, and examining risks the system did not flag.
The important problem is not necessarily a false output.
It is the incomplete identification of material issues.
Legal review must therefore examine both what AI identifies and what it overlooks.
The Litigation Prediction
A litigation attorney uses an AI analytics platform to estimate potential case outcomes.
The platform generates a probability estimate, but the available information does not adequately establish the dataset’s scope, relevant assumptions, or comparability to the client’s matter.
The attorney must assess the prediction’s limitations before using it as support for legal advice.
Relevant considerations may include case-specific facts, jurisdiction, procedural posture, available evidence, client objectives, and competing strategies.
The AI estimate may be a useful input into analysis.
It should not be presented as an established legal outcome without sufficient support.
These examples demonstrate that defensible verification is not a single mechanical checklist.
The relevant professional inquiry must be adapted to the nature of the work.
From Individual Verification to Institutional Accountability
Individual professional judgment remains essential, but law firms should also consider how verification is implemented across teams.
A policy requiring attorneys to review AI-generated work is useful only if the organization establishes an appropriate understanding of what review means for different tasks.
Firms can support accountability through tool approval processes, role-specific training, task-based verification, escalation procedures, and periodic review of recurring errors.
Supervisory attorneys should understand when additional review is required and which decisions remain reserved for qualified lawyers.
These procedures should be proportionate to the firm’s practice areas, technology, and client needs.
The broader institutional challenge is ensuring that formal policies correspond to actual operating practices.
That issue is explored in Building an AI Governance Framework for Law Firms: Six Pillars of Operational Accountability .
Verification is one critical component of that larger governance structure.
A firm must preserve professional responsibility not only in its written commitments but also in the workflows through which legal work is performed.
The Bottom Line: Defensibility Begins Before the Filing
Artificial intelligence can accelerate research, organize information, generate preliminary analysis, and assist lawyers with increasingly complex work.
But those capabilities do not eliminate the need for independent professional verification.
The central lesson of the emerging judicial decisions is not simply that hallucinated citations can lead to sanctions.
It is that lawyers remain responsible for the authenticity, accuracy, and appropriate use of the authorities and representations they submit.
The same professional principle matters in client counseling, contract review, transaction analysis, litigation strategy, and other AI-assisted legal services.
Source integrity establishes the foundation.
Substantive verification tests whether the legal reasoning supports the conclusion.
Process accountability clarifies how the work was reviewed and who assumed responsibility for consequential decisions.
“The question is not simply whether AI generated an accurate answer. It is whether the lawyer understood the question, evaluated the evidence, tested the reasoning, and made a defensible decision about relying on the result.”
Jeff Howell, Esq. | Lex Wire Journal, Professional Analysis
The Defensible AI Work Product Framework offers a practical way to organize those responsibilities.
It does not require every matter to follow an identical procedure.
It requires the lawyer or firm to consider what appropriate verification demands in the circumstances.
The professional standard is not satisfied merely because an AI output looks correct. Defensible legal work requires a reasonable basis for trusting the evidence, reasoning, and professional judgment behind the result.
Related Lex Wire Journal Analysis
- The Ethics of Artificial Intelligence in Legal Practice: Professional Judgment in the Age of Automation examines how attorneys preserve independent judgment when AI performs substantive legal work.
- Client Communication in the Age of AI addresses transparency, informed consent, Rule 1.4, and client decision-making.
- AI and the Attorney-Client Privilege explores confidentiality, third-party platforms, work product, and information control.
- Building an AI Governance Framework for Law Firms develops the institutional foundations of operational accountability.
- Building Your Law Firm’s AI Playbook addresses practical AI procedures, training, and implementation.
Legal Authorities and Source Materials
Judicial Decisions
- Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023). Opinion and Order on Sanctions .
- LNU v. Blanche, No. 24-4790 (9th Cir. June 3, 2026). Published Disciplinary Order .
- LeDoux v. Outliers, Inc., No. 3:24-cv-05808-TMC (W.D. Wash. July 24, 2026). Order on Sanctions .
Professional Responsibility and Procedural Rules
- American Bar Association, Formal Opinion 512: Generative Artificial Intelligence Tools (July 29, 2024).
- American Bar Association, Model Rules of Professional Conduct , including Rules 1.1, 3.3, 5.1, and 5.3.
- Federal Rule of Civil Procedure 11 , representations to the court and sanctions.
Original CLE Instructional Materials
- Jeff Howell, Esq., Ethics in the Age of AI: Presentation Outline and Script , continuing legal education instructional materials, sections addressing AI-assisted research, accuracy verification, challenge protocols, supervision, and responsible implementation.