Can I use a free chatbot for legal research?

For research into real cases and statutes, no, not as a source of truth. General chatbots can invent authorities that sound plausible and do not exist, which is exactly how lawyers have been sanctioned. You can use one to help you think or to draft, but every case, statute or rule must be confirmed in a real legal database before you rely on it.

Is it a confidentiality breach to paste a client document into AI?

It can be. Entering privileged or confidential client information into a general consumer tool may expose it to a third party and, depending on the terms, feed it into training. ABA Formal Opinion 512 treats this under the confidentiality duty in Model Rule 1.6 and points toward client consent and vetted tools. Check the tool's data terms and your own regulator's rules before entering anything client-identifying.

What actually happened in the Mata v. Avianca case?

Two lawyers filed a brief in the Southern District of New York citing cases that a chatbot had fabricated, including fake quotations. Nobody could find the cases because they did not exist. In June 2023, Judge P. Kevin Castel sanctioned the lawyers 5,000 dollars. It became the standard warning, and courts have sanctioned other lawyers for the same error since.

Will using AI get me in trouble with the bar?

Using AI is not itself a violation. Filing fabricated citations, breaching client confidentiality or failing to supervise the work is. ABA Formal Opinion 512, issued in July 2024, applies existing duties rather than creating new prohibitions. The risk comes from skipping verification, not from the tool. Confirm your specific obligations with your state bar, as rules differ by jurisdiction.

Which tasks are the safest place to start?

Start where you already hold the source material and are checking output against it: summarising a contract you have, drafting a client email, turning your own analysis into plain language, or organising intake answers. These save real time at low risk, provided a lawyer reviews the result before it leaves the office. Keep anything that gets filed or relied on under strict verification.

Do I need to tell clients I am using AI?

Possibly, and it is worth being transparent regardless. Guidance and rules on disclosure and consent are still developing, and ABA Formal Opinion 512 discusses communication and informed consent, particularly where client information goes into a tool. Boilerplate consent buried in an engagement letter may not be enough. Check your jurisdiction's current position and, when unsure, ask the client directly.

Explainer

AI for small law firms: what is safe, and what is not?

September 29, 2026 · 7 minute read

The short answer

AI is safe for a small law firm when it drafts, summarises and explains, and a lawyer checks every output before it leaves the building. It is not safe when you file its work unverified, or when you paste privileged or confidential client information into a general consumer tool you have not vetted. Those two lines are where lawyers have been sanctioned, and everything else follows from them. This article is general information, not legal or ethics advice, so confirm your own obligations against your state bar before you set firm policy.

The line that actually matters

There is a lot of noise about AI in law, most of it either breathless or frightened. Strip it back and the safe-or-unsafe question has a boringly practical answer. AI is a fast, confident, sometimes wrong assistant. Treat everything it produces as a draft from a keen junior whose work has never been checked, and you will be fine. Treat it as a finished source of truth, and you will eventually file something that does not exist.

Two rules cover almost every real risk. Never rely on AI output, in a filing or in advice, without independently verifying it. Never put privileged or confidential client information into a general consumer tool whose terms you have not read. Everything below is a consequence of those two sentences.

What went wrong, and why it keeps happening

The cautionary tale is real and worth knowing precisely. In Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), two lawyers submitted a brief citing cases that a chatbot had invented, complete with fabricated quotations. When opposing counsel could not find the cases, neither could the court. Judge P. Kevin Castel sanctioned the lawyers 5,000 dollars in June 2023, and the matter became the standard warning.

It did not stop there. In February 2025, a federal court in Wyoming sanctioned three attorneys at Morgan and Morgan, one of the largest plaintiffs' firms in the country, after a filing cited cases that did not exist. The pattern is always the same: the tool sounded authoritative, nobody checked the citations, and the fabrications reached a judge. This is a verification failure, not an AI failure.

The lesson is not to avoid AI. It is that a confident tone is not evidence. A citation is only real when you have pulled the actual authority from a real database and read it.

What is genuinely safe to speed up

Plenty of daily work is low-risk and a good fit, as long as a person signs off before anything goes out. The safe zone is drafting, summarising and explaining, where you already hold the source material and you are checking the output against it.

The common thread: the AI works from information you supplied or can immediately verify, and a qualified person reviews the result. Nothing here is filed or acted on unchecked.

Lower-risk tasks worth trying

First drafts of client emails, engagement letters and standard clauses, then edited to fit the matter
Summarising a document you already hold, such as a long contract or a deposition transcript, with the summary checked against the source
Turning your own dense analysis into a plain-language explanation for a client
Intake work: drafting questionnaires, organising a client's answers, spotting missing information
Rewriting your own prose to be shorter or clearer, where you can see exactly what changed

What stays off-limits

Some uses carry risk out of proportion to the time they save. Treat these as off-limits in any general consumer tool, and approach them carefully even in a paid legal-specific product.

The confidentiality point deserves emphasis. Pasting client facts, names or documents into a free consumer chatbot can expose privileged information to a third party and, depending on the tool, feed it into future training. That is a professional-responsibility problem, not just a technical one.

Do not do these

Filing or citing anything from AI without pulling and reading the actual authority yourself
Entering privileged or confidential client information into a general consumer tool you have not vetted
Asking a general chatbot for legal research and treating its cases or statutes as real
Letting AI give the client advice directly, without a lawyer's review
Using AI output to make a decision you could not defend if a judge asked how you reached it

Building the verification habit

The firms that use AI without incident are not more cautious by temperament. They have a fixed process. Verification is a step in the workflow, not a matter of remembering to be careful on a busy Friday.

A simple discipline covers most of the exposure. Write it down, and hold everyone in the practice to it, yourself included.

A minimum standard

Every case, statute or rule an AI mentions is confirmed in a real legal database before it appears anywhere
Every factual claim is traced back to a source you can see
Client-identifying information is stripped out, or the tool is one your firm has formally approved for confidential data
A named person reviews AI-assisted work before it goes to a client or a court
You can explain, if asked, exactly what the AI did and what you checked

Choosing tools and setting a policy

Tools differ most on the point that matters: what happens to what you type in. Before adopting anything, read its terms on data retention and training. Prefer tools that let you switch training off, or that are built for legal work with contractual confidentiality protections. A paid business tier is usually clearer on this than a free consumer login.

The American Bar Association issued Formal Opinion 512 on 29 July 2024, its first ethics guidance on generative AI. It does not ban the technology. It applies existing duties, including competence under Model Rule 1.1, confidentiality under Model Rule 1.6, candour to the tribunal and supervision, to how you use it. The opinion also treats boilerplate consent buried in an engagement letter as unlikely to be enough where client information goes into a tool. Reading it is an hour well spent for any practice owner.

This article is general information, not legal or ethics advice. Rules vary by state and by regulator, and they are moving quickly. Confirm your own obligations against your state bar's guidance and your professional-liability policy before you set firm policy.

Common questions

Can I use a free chatbot for legal research?: For research into real cases and statutes, no, not as a source of truth. General chatbots can invent authorities that sound plausible and do not exist, which is exactly how lawyers have been sanctioned. You can use one to help you think or to draft, but every case, statute or rule must be confirmed in a real legal database before you rely on it.
Is it a confidentiality breach to paste a client document into AI?: It can be. Entering privileged or confidential client information into a general consumer tool may expose it to a third party and, depending on the terms, feed it into training. ABA Formal Opinion 512 treats this under the confidentiality duty in Model Rule 1.6 and points toward client consent and vetted tools. Check the tool's data terms and your own regulator's rules before entering anything client-identifying.
What actually happened in the Mata v. Avianca case?: Two lawyers filed a brief in the Southern District of New York citing cases that a chatbot had fabricated, including fake quotations. Nobody could find the cases because they did not exist. In June 2023, Judge P. Kevin Castel sanctioned the lawyers 5,000 dollars. It became the standard warning, and courts have sanctioned other lawyers for the same error since.
Will using AI get me in trouble with the bar?: Using AI is not itself a violation. Filing fabricated citations, breaching client confidentiality or failing to supervise the work is. ABA Formal Opinion 512, issued in July 2024, applies existing duties rather than creating new prohibitions. The risk comes from skipping verification, not from the tool. Confirm your specific obligations with your state bar, as rules differ by jurisdiction.
Which tasks are the safest place to start?: Start where you already hold the source material and are checking output against it: summarising a contract you have, drafting a client email, turning your own analysis into plain language, or organising intake answers. These save real time at low risk, provided a lawyer reviews the result before it leaves the office. Keep anything that gets filed or relied on under strict verification.
Do I need to tell clients I am using AI?: Possibly, and it is worth being transparent regardless. Guidance and rules on disclosure and consent are still developing, and ABA Formal Opinion 512 discusses communication and informed consent, particularly where client information goes into a tool. Boilerplate consent buried in an engagement letter may not be enough. Check your jurisdiction's current position and, when unsure, ask the client directly.