How to Study AI in Court Filings for the MPRE

Mata v. Avianca is the cleanest way to study AI in court filings for a Professional Responsibility exam because the facts already look like an issue-spotter. A lawyer used ChatGPT while preparing a court filing. The filing cited legal authorities that did not exist. The problem reached the court, and sanctions followed: a $5,000 fine and a finding of subjective bad faith in a 2023 Southern District of New York decision. [1]

Do not memorize that case as “AI made up cases.” That is the newspaper version. The exam version is narrower and more useful: a lawyer put unverified legal material into a tribunal filing, then had to answer under duties that already governed lawyer work before anyone used a chatbot.

ABA Formal Opinion 512, issued on July 29, 2024, makes the same move. It does not create a separate ethics code for “AI lawyering.” It applies existing Model Rules to generative AI use, including competence, communication, fees, confidentiality, candor, supervision, and misconduct. [2] For MPRE purposes, that is the controlling study frame: Model Rule first, AI tool second.

The fast rule map

If a fact pattern says a lawyer used AI to draft, research, summarize, cite-check, translate, or file something, resist the urge to invent an “AI rule.” Ask what lawyer duty was already in play.

Rule 1.1: competence is where the AI issue usually starts

Rule 1.1 is the first stop because the lawyer chose the method of work. If a lawyer uses AI to generate legal research, draft a brief section, summarize a record, or propose citations, the lawyer still has to provide competent representation. Formal Opinion 512 treats that as an application of the existing competence duty, including the need to understand the benefits and risks of the technology used in the representation. [2]

On an exam, competence is not tested by asking whether AI is impressive. It is tested by asking whether the lawyer’s use of the tool was reasonable for the task and whether the lawyer checked the output before relying on it. A draft that saves time is still only a draft. A citation that appears in polished Bluebook form is still only a claim until it has been verified against actual legal authority.

Mata makes that point concrete. The ethics problem was not merely that ChatGPT produced fabricated authorities. The sanctionable conduct arose when those authorities were submitted in a real court filing and the lawyers failed to perform the verification that legal research has always required. The court imposed a $5,000 sanction and found subjective bad faith. [1]

That is why “the software made the mistake” is not a complete answer. It may explain the source of the false material, but it does not answer the Rule 1.1 question. The lawyer controlled whether the material entered the filing. The lawyer controlled whether the cases were checked. The lawyer controlled whether a court received unverified authority as though it were real law.

What to eliminate in an MPRE answer set

• Eliminate any answer saying AI use is automatically unethical. The materials support a narrower rule: irresponsible or unverified use can violate existing duties.
• Eliminate any answer saying the lawyer is safe because the tool generated the mistake. Competence is not delegated to software.
• Be suspicious of answers that treat a polished AI response as a substitute for legal research. The duty is to verify before use.

Rule 3.3: fabricated citations move the problem from bad research to candor

Rule 3.3 is the rule to see when the false material reaches a tribunal. A hallucinated case in a private draft may begin as a competence problem. A hallucinated case cited to a court as real authority raises candor to the tribunal.

For exam purposes, keep the line clean. Rule 1.1 asks whether the lawyer handled the work competently. Rule 3.3 asks whether the lawyer made, failed to correct, or allowed a false statement to the tribunal. AI is just the mechanism by which the false material entered the lawyer’s workspace.

Mata is useful because the authorities were not merely weak, distinguishable, or overread. They were fabricated. That makes the case memorable, but the rule does not depend on the technology being exotic. A lawyer who inserts nonexistent authority into a filing has created the same kind of tribunal-facing problem whether the false citation came from a chatbot, a sloppy associate memo, an old form file, or the lawyer’s own notes.

The exam move is to ask when the lawyer knew or should have confronted the falsity. Once the filing is challenged, the lawyer cannot hide inside uncertainty. The duties of candor and correction are triggered by the lawyer’s role as an officer of the court, not by the marketing label on the research tool.

Competence and candor often appear together, but they are not the same issue

A good answer can mention both rules if the facts support both. The lawyer may have been incompetent in failing to verify the AI-generated research and may also have violated candor duties by submitting or failing to correct false authorities in court. Do not collapse them into one vague “AI ethics” violation. The MPRE rewards the narrower duty.

Rules 5.1 and 5.3: supervision is the escape route that usually fails

AI fact patterns often add a layer between the lawyer and the error. A junior associate used the tool. A paralegal entered the query. A contractor generated a first draft. A firm workflow routed AI summaries into a brief bank. That extra layer matters for spotting supervision, but it does not erase the underlying lawyer duty.

Rule 5.1 deals with responsibilities regarding other lawyers. Rule 5.3 deals with responsibilities regarding nonlawyer assistance. Formal Opinion 512 places AI use within that existing supervisory frame: lawyers with managerial or supervisory authority must make reasonable efforts to ensure that the use of generative AI is compatible with professional obligations. [2]

Here is the exam trap: the answer choice says the partner is not responsible because the junior lawyer personally generated the draft. That is too quick. The signing lawyer or supervising lawyer may still have duties to review the work, set reasonable procedures, and prevent unverified AI output from entering a filing.

The same analysis applies when the person using the tool is not a lawyer. A paralegal, clerk, outsourced vendor, or document-services provider can create the factual path, but the professional duty remains with the lawyer. The question is not whether the nonlawyer violated the Model Rules. The question is whether the lawyer adequately directed, supervised, and reviewed the work.

• If the facts emphasize a signed filing, look for the lawyer’s own competence and candor duties.
• If the facts emphasize a subordinate’s AI use, add Rule 5.1.
• If the facts emphasize a paralegal, vendor, staff member, or tool workflow, add Rule 5.3.
• If the facts emphasize firm policy, training, or review systems, supervision is probably not decoration. It is the issue.

Rule 1.6: confidentiality is the upload problem

Confidentiality appears when the lawyer puts client information into an AI system or allows client information to be processed through a tool without adequate protection. Formal Opinion 512 identifies Rule 1.6 as one of the existing duties implicated by generative AI use. [2]

The exam fact pattern may not say “court filing” at this stage. It may say the lawyer uploaded a draft complaint, medical records, settlement facts, client emails, or a confidential litigation strategy to get a summary. If the information relates to the representation and the lawyer has not taken appropriate steps to protect it, the issue is confidentiality.

Do not overstate the rule. The mere use of technology is not automatically a confidentiality violation. The better question is whether the lawyer took reasonable precautions for the sensitivity of the information, the nature of the tool, and the representation.

Rule 1.4: communication matters when AI use affects the client’s decision

Rule 1.4 is not a magic notice requirement for every spell-check, search query, or drafting aid. It becomes important when the AI use is material to the representation or to a decision the client must make. Formal Opinion 512 treats client communication as part of the existing Model Rule analysis for generative AI. [2]

A fact pattern may make communication relevant by tying AI use to cost, confidentiality, litigation strategy, delegation, or the reliability of an important filing. If the client needs information to make an informed decision, the lawyer may need to explain the use, limits, or risks of the tool in a way the client can understand.

Rule 1.5: AI can make the billing fact pattern testable

Fees become an issue when AI-assisted work is billed in a way that misrepresents time, value, or the basis of the charge. Formal Opinion 512 includes fees among the existing duties lawyers must consider when using generative AI. [2]

The clean exam version is not “AI made the work cheaper, so the lawyer must charge less.” That is too broad. Look instead for unreasonable fees, double billing, billing for time not actually spent, or descriptions that make the client think a lawyer performed work that was actually generated automatically and barely reviewed. The misconduct may be in the billing representation, not in the use of the tool itself.

Rule 8.4: when the AI mistake becomes dishonesty

Rule 8.4 is the broader misconduct backstop. It becomes more likely when the lawyer lies, conceals, fabricates, or acts dishonestly after the AI-generated problem is exposed. Formal Opinion 512 includes misconduct among the existing Model Rule duties implicated by generative AI. [2]

This is where a fact pattern may shift from “bad verification” to “dishonest response.” A lawyer who discovers that a cited case may not exist and then misleads the court about the source or reliability of the citation has created a different problem from the original research failure. The exam answer should track that change.

Why verification is the operational skill

The rule map keeps returning to verification because AI-generated legal material can look finished before it has earned trust. A model can produce a case name, reporter citation, parenthetical, quotation, or procedural history in the visual style of legal research. That appearance is not authority.

For study purposes, do not make the verification rule complicated. Before legal material goes into a filing, the lawyer must confirm that the authority exists, says what the filing claims it says, remains good law for the proposition used, and fits the procedural posture. If the AI output summarizes facts or a record, the lawyer must check it against the actual record. If the output quotes a source, the lawyer must check the quotation.

This is also why a court-filing fact pattern is more dangerous than a private brainstorming fact pattern. A private draft still requires care, but a filed document reaches the tribunal and may affect an opponent, a client, and judicial resources. Once the document is filed, competence and candor can sit on the same set of facts.

An MPRE spotting routine for AI court-filing questions

When an exam question includes AI, move in this order. The goal is not to sound current. The goal is to find the duty that the lawyer controlled.

1. Identify the AI-assisted act. Was the tool used for research, drafting, citation, record review, client communication, billing, translation, or document handling?
2. Locate the lawyer-controlled duty. Filing with a court points to competence and candor. Client information points to confidentiality. Subordinate or vendor use points to supervision. Client decision-making points to communication. Billing points to fees.
3. Ask whether the output was verified before use. If the output supplied law, facts, quotes, citations, or record summaries, verification is the central conduct question.
4. Separate the first mistake from the later response. A careless unverified draft may be competence. A false filing may be candor. A cover-up or misleading explanation may add misconduct.
5. Check who supervised the workflow. If a junior lawyer, staff member, contractor, or vendor used the tool, ask what the supervising lawyer did to prevent, catch, or correct the problem.

Here is the same routine as a margin-note checklist:

Mata should stay in your memory as the anchor case: AI-generated citations, fabricated authorities, a real court filing, sanctions, a $5,000 fine, and subjective bad faith. [1] But the deeper lesson is the ordinary one. AI in court filings changes the fact pattern, not the ethical architecture.

References

1. Mata v. Avianca, Inc., 678 F. Supp. 3d 443 — CourtListener — 2023 — https://www.courtlistener.com/opinion/9424564/mata-v-avianca-inc/
2. Formal Opinion 512: Generative Artificial Intelligence Tools — American Bar Association — July 29, 2024 — https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf