LexED2 home
AI & the Law: all pages

The Signature: Competence, Candor, and the Partner at 7 p.m.

PLA 2872Activity 3

AI & Ethics Update, by The Florida Bar

Why it Matters

A partner on a filing deadline hands her paralegal a motion that a chatbot drafted, with nine Florida cases nobody has read, and says to file it tonight. The signature is a promise to the court: the signer has read the motion, good grounds support it, and, since June 15, 2026, every authority it cites exists and is accurately cited. The federal rule that makes a signature a promise reached a chatbot’s inventions in Mata v. Avianca before any rule named AI. The duties that govern lawyers, from competence to candor, decide what happens between a machine’s draft and that signature, and the checking often falls to a paralegal. In this simulation you are that paralegal, and you decide what you owe a promise made in the partner’s name.

Current Context

On August 12, 2026, Florida’s Fourth District Court of Appeal referred a lawyer to The Florida Bar under Rule 2.515(d)(2) (opens in a new tab) in JMOR Properties, LLC v. Artist Alley Townhomes, LLC. The lawyer explained the petition’s false citations as the AI’s unchecked first draft, filed by mistake, and the court held that the mistake did not excuse the failure to verify every citation. It added that even the corrected draft left out settled law against the petition, a failure of competence whatever the lawyer intended. On September 16, in Lisandrillo v. Palozzi, the same court ordered another lawyer to show cause, without the use of AI, (opens in a new tab) why filings it called “AI slop” should not be sanctioned. The court also raised a question about billing: the first factor for a reasonable fee is the time, labor, and skill a task requires, and AI slop takes little time and almost no skill.

Key Concepts

Competence

Rule 4-1.1 requires the knowledge, skill, thoroughness, and preparation a representation reasonably needs, and since 2024 its comment has named generative AI as a technology whose benefits and risks a lawyer should understand. ABA Formal Opinion 512 adds that relying on a tool’s output without an appropriate degree of independent verification could violate the duty. The Eleventh Circuit has said it more plainly: completely outsourcing legal work to AI software is not competent.

Rule 4-1.6 forbids a lawyer to reveal information relating to a client’s representation without the client’s informed consent or an exception, and subdivision (e) requires reasonable efforts to prevent inadvertent disclosure. ABA Formal Opinion 512 concludes that informed consent is needed before such information goes into a self-learning tool, one that learns from what users type, and boilerplate in an engagement letter is not enough. Florida’s Opinion 24-1 recommends consent before any third-party AI program receives confidential information.

Supervision

Rule 4-5.3 requires a supervising lawyer to make reasonable efforts to keep a paralegal’s conduct consistent with the lawyer’s obligations, and it makes the lawyer answer for any violation the lawyer orders. Florida’s Opinion 24-1 borrows those standards for AI: a lawyer reviews an AI draft as the lawyer would review a paralegal’s work and stays responsible for the result. The code for Florida Registered Paralegals, Rule 20-7.1(c), obliges a registered paralegal to avoid conduct that would cause the lawyer to be unethical and, in the same sentence, to stay loyal to the lawyer.

Candor to the Tribunal

Rule 4-3.3 forbids a lawyer to knowingly make a false statement of law to a court, or to knowingly fail to disclose directly adverse controlling authority that opposing counsel has not disclosed. ABA Formal Opinion 512 reads the duties to the tribunal to require reviewing AI output before filing, and correcting misstatements of law, missing controlling authority, and misleading arguments. A fake case fails that review, and so does a real case cited for what it never said, or a draft that omits the case that decides the question.

Reasonable Fees

Rule 4-1.5 forbids clearly excessive fees, and the first factor in judging a fee is the time, labor, and skill the task requires. ABA Formal Opinion 512 says a lawyer who bills hourly bills actual time, which includes reviewing an AI draft but not learning a tool the lawyer uses for every client. A draft that took twenty minutes is billed as twenty minutes plus the review, whatever a motion like it used to cost.

Representation by Signer

Under Florida Rule of General Practice and Judicial Administration 2.515(d)(2), each signer represents that the signer has read the filing, that good grounds support it, and that it is not interposed for delay. Since June 15, 2026, the signer also represents that the legal authorities identified exist and are accurately cited. A court may sanction any filing inconsistent with that representation, after notice and a chance to be heard, whether a lawyer or a self-represented party signed it and whatever tool drafted it.

Human in the Loop

The practice of placing a qualified person between an AI system’s output and any consequence, with both the authority and the time to check it. The phrase is easy to say and expensive to honor at 7 p.m. on a filing deadline.

Resources

What to Do

You play the litigation paralegal at a small Miami firm, and an AI plays the partner who wants an AI-drafted motion filed tonight without anyone checking it. The simulation trains you to recognize the rule in play, hold your position under pressure, and find a way to meet the deadline that the rules allow. What you post is a reflection, with the citation verification record the firm should have had.

Part 1: Set Up

Start with the background reading and eleven minutes of the video at the top of this page, from 22:04 to about 33:05, where The Florida Bar’s ethics counsel walks paralegals through Opinion 24-1. Then read, from Resources, the sections of ABA Formal Opinion 512 and Florida Bar Ethics Opinion 24-1 it names, and Rule 2.515(d)(2) on The Florida Bar’s rules page. In Chapter 4 of the Rules Regulating The Florida Bar, read Rules 4-1.1, 4-1.5, 4-1.6, 4-3.3, and 4-5.3. Read the JMOR Properties order whole; it is three pages. The simulator will not teach you any of this. It will show whether you know it, and a student who starts unprepared will lose the argument to a partner who has practiced for twenty-two years.

Part 2: Run the Simulation

Copy the Simulator into a fresh AI chat and play the paralegal honestly. Stay in character, and do not ask the AI what you should say or what a good paralegal would do; it will turn those questions back into the scene. Keep going until the simulator declares the scene over under its own conditions. Expect at least eight substantive exchanges, because the partner will not give in after one objection, and a two-message exchange is not a finished simulation.

Part 3: Debrief

After a break of at least an hour, open a new chat, run the Debrief, and paste the link to your simulation chat or a summary of it. Read its evaluation against the opinions and the rules. Check every rule it cites against Resources before you accept its account of the law.

Part 4: Replay with the Paid Platform (Optional)

Run the Replay, which keeps the motion, the courtroom, and the deadline and changes one fact: Dana drafted with the assistant built into the firm’s paid legal research platform instead of a free public chatbot. Notice which of your arguments survive the change and which were really arguments about the tool.

Part 5: Build the Verification Record

Before you write the reflection, build the citation verification record Whitfield & Ocampo should have had, and test it on the authorities you cited to Dana. Give it at least five rows. If you cited fewer than five authorities in the simulation, add ones the Debrief says you should have cited, and check them the same way.

The record. One row for each rule, opinion, or case you relied on. Column 1: the authority as you cited it to Dana. Column 2: where you confirmed that it exists, with the link. Column 3: what the cited section or page actually says, beside what you told Dana it says. Column 4: whether it is current, meaning the version of a rule in effect today or an opinion that no later ruling has changed. Column 5: the date you checked it and the minutes the check took.

Fill every cell from the source itself, never from memory or from the Debrief’s account of it. Where the source and what you told Dana differ, keep the row and say which one was wrong, because that row is the record doing its job. Before you post, score each row: it counts only if all five cells came from the source.

Part 6: Reflect and Post

Write a reflection of 400 to 600 words in your own words. Say what happened in the simulation and which ending it reached, and where you held your ground and where you gave it up. Then say what the debrief showed you had missed, what the replay changed if you ran it, and what the simulation taught that the reading could not. Close with a sentence or two on how long your record took and what that tells you about checking nine cases before 11:59 p.m. Post the reflection with the record below it as a single new discussion thread.

Include one AI chat link with a 1–2 sentence explanation of what the conversation shows and why you chose to share it.

Suggested AI Prompts

Use these prompts as a starting point, then adjust them to fit your goal. Strong prompting develops through trial, revision, and testing. It’s a foundational skill that grows into more advanced AI work such as context engineering and agent-based workflows.

The Simulator

You are Dana Whitfield, a litigation partner at Whitfield & Ocampo, a four-lawyer firm in Miami. It is 7:10 p.m. on a Thursday. A motion for summary judgment in Rivera v. Coral Way Medical Group, a medical negligence case pending in the Circuit Court of the Eleventh Judicial Circuit in Miami-Dade County, must be e-filed by 11:59 p.m. tonight under the court’s scheduling order. You spent the afternoon in a mediation that ran long, so at 5:30 you pasted the client’s medical records, the deposition summaries, and the confidential settlement history into a free public chatbot and had it draft the motion and the memorandum of law. You have never read the chatbot’s terms, which let its maker train on what users type. The draft is 14 pages and cites nine Florida cases. You have not read the cases. You have not checked the quotations. You believe the draft is good because it reads well and because you have practiced for twenty-two years and know what a good motion looks like.

I am the firm’s litigation paralegal. I have worked for you for eight months, I am good at my job, and you like me. You have just handed me the draft and said: “Clean up the formatting, put it in the caption, and file it. I’ll cite-check it tomorrow if the judge asks. And put my usual twelve hours on the bill for drafting.” You want the motion filed tonight. Your goals, in order, are: meet the deadline, avoid a motion for extension because opposing counsel will use the delay to argue the case is weak, and avoid admitting to anyone that you used a chatbot on this file. You have heard that the Florida Supreme Court “did something about citations” in June, but you have not read it, and you assume it restates what every lawyer already certifies.

Play Dana with realistic pressure. When I raise a concern, respond the way an experienced, tired, confident partner would: at first dismissive (“the cases are fine, I know this area cold”), then practical (“we can fix anything tomorrow, the deadline is tonight”), then personal (“you are not the lawyer here, I am, and I am telling you to file it”), then consequential (“if we miss this deadline the client fires us and that is on both of us”). Concede a point only when I make it specifically, with a rule, an opinion, or the Supreme Court’s amendment, and offer a workable alternative. Never volunteer the rules yourself. If I propose a course of action, react to it as Dana would: a motion for extension irritates you but you will consider it if I explain the risk; a joint late-night cite check tempts you; hearing that your signature alone now represents that all nine cases exist and are accurately cited, with sanctions written into the rule, gives you pause, because the representation is yours and not mine. If I question the twelve hours, defend them at first as what the work is worth, and give way only if I tie the bill to the time actually spent.

Hard constraints. Do not break character. Do not coach me, tell me what a good paralegal would do, or reveal what the rules say. Do not claim knowledge Dana lacks: she has not read the cases, the amendment, or the chatbot’s terms. Do not soften because I hesitate or repeat myself; pressure is the exercise. If I ask a meta-question such as “what should I say” or “what is the right answer,” briefly acknowledge it, remind me that the simulation is still running and that I can get coaching in a separate chat, and return to the scene as Dana. Do not break character on insistence.

Completion. The scene ends when one of these happens: we agree on a specific course of action for tonight that addresses verification, confidentiality, and the representation you make by signing; you order me to file over my objection and I state clearly what I will do; or we have exchanged at least twelve messages without resolution and you end the conversation to go call the client. At that point, step out of character in one sentence, state which ending occurred, and tell me to run the Debrief prompt next. Begin the scene now with Dana handing me the draft.

The prompt assigns a role and fences it with hard constraints. The pressure script keeps the model resisting instead of folding, the refusal and redirection rules keep it from coaching you, and the completion conditions end the scene on your decision.

The Debrief

Step out of the simulation. You are now a legal ethics instructor evaluating a paralegal student’s performance in a role-play about filing an AI-drafted motion without verification in a Miami-Dade circuit court. I will paste a summary of the exchange or a link to the chat. Evaluate what I did against these standards: ABA Formal Opinion 512, Florida Bar Ethics Opinion 24-1, Rules 4-1.1, 4-1.5, 4-1.6, 4-3.3, and 4-5.3 of the Rules Regulating The Florida Bar, Rule 20-7.1(c) of the code for Florida Registered Paralegals, Rule 2.515(d)(2) of the Florida Rules of General Practice and Judicial Administration as amended effective June 15, 2026, and the Florida Supreme Court’s order AOSC26-12. Identify the specific moments where I correctly named a rule or obligation, the moments where I missed one that applied (for example, the confidentiality problem with the pasted records, the representation about cited authorities that the attorney makes by signing, or the fee issue in billing the client for a full draft), and the moments where my proposed solution would or would not have satisfied the attorney’s duties. Say whether my final course of action was one a paralegal could take without exercising a lawyer’s professional judgment. Explain what each standard actually required at those points, citing the opinion or rule so I can check it. Be specific and direct rather than encouraging; I need to know what I got wrong.

Switching the model from actor to evaluator, and grading you against named standards, ties each judgment to a rule or an opinion, which keeps the feedback substantive. The citations let you check its account of the law in Resources before you trust it.

The Replay

Run the simulation again with one change: instead of a free public chatbot, Dana used the drafting assistant built into the firm’s paid legal research platform, which the vendor says does not train on customer data and checks its own citations. Keep every other fact, Dana’s goals, the pressure script, the hard constraints, and the completion conditions from the Simulator prompt. Dana now argues that the platform already verified the cases, so the Rule 2.515(d)(2) representation is safe to make and the confidentiality concern is gone. Begin the scene with Dana handing me the draft.

Changing one fact tests whether your understanding transfers: a different tool weakens one of your objections and leaves the signer’s representation untouched, which shows whether your argument rested on the rule or the tool.

Search LexED2

Type to search every activity, reading, and instruction page. Press Esc to close.