Law firms hold some of the most private information people ever share. Divorce details, medical records, business secrets, and criminal histories all pass through a lawyer’s office. Now many firms use AI tools to draft documents, summarize files, and answer research questions. These tools save time. But they also create new ways for client data to leak. A paralegal might paste a custody agreement into a free chatbot. A vendor might store prompts on servers the firm never checked. And once that data is out, you can’t pull it back. Bar associations have noticed. The ABA and state bars, including The Florida Bar, expect lawyers to understand how AI tools handle data before they use them. None of this means you should avoid AI. This article explains the main risks, what the ethics rules say, and the practical steps your firm can take to protect client information while still getting value from AI.

Where AI Tools Put Client Data at Risk

Most AI tools work by sending what you type to a server run by someone else. That’s the core issue. Your firm loses some control the moment a prompt leaves your network.

Here are the risks that come up most often:

• Training on your inputs. Some consumer AI tools use prompts to improve their models unless you opt out. Client facts could, in theory, shape future outputs for other users.
• Data retention. Many vendors keep prompts and uploaded files for 30 days or longer. Some keep them for abuse monitoring even on paid plans.
• Shadow AI. Staff use personal accounts on free tools because the firm hasn’t given them an approved option. The firm has no record of what was shared.
• Third-party access. AI vendors often rely on cloud providers and subcontractors. Each one is another party that could be breached.
• Prompt injection and bad outputs. A document you upload may contain hidden instructions that change how the tool behaves. And AI can produce made-up case citations, which is a separate but real risk to clients.

The common thread is simple. If you don’t know where the data goes, you can’t promise a client it’s safe.

What the Ethics Rules Expect From Lawyers

There’s no special AI rulebook for lawyers. The existing rules already apply, and bar groups have explained how.

ABA Formal Opinion 512 (July 2024) was the ABA’s first formal opinion on generative AI. It says the duty of competence includes knowing how these tools work. It also says lawyers need informed client consent before putting information about a case into a self-learning AI tool. The ABA notes that consent doesn’t have to be written, but confirming it in writing is the safer practice.

Florida Bar Ethics Opinion 24-1 (January 2024) says Florida lawyers may use generative AI if they protect confidentiality, supervise the work, keep fees reasonable, and follow advertising rules. It recommends getting the client’s informed consent before using a third-party AI program that would see confidential information. It also says in-house AI, where no outside party hosts or stores the data, can reduce that concern.

Other states, including California, New York, and Texas, have issued similar guidance. The rules they lean on are mostly the same:

• Rule 1.1 (Competence)
• Rule 1.6 (Confidentiality of Information)
• Rule 1.5 (Fees)
• Rule 5.3 (Supervising Nonlawyer Assistance)

So the question isn’t whether you can use AI. It’s whether you can show you took reasonable steps to protect client information when you did.

Questions to Ask Before You Sign With an AI Vendor

The Florida Bar’s guidance on cloud tools gives a good starting point. Make sure the vendor has to keep your data confidential, that the promise is enforceable, and that the vendor will tell you if someone demands your data.

Before you approve any AI tool, get clear answers to these questions:

1. Does the vendor use our prompts or files to train its models? Can we turn that off in the contract, not just in settings?
2. How long are prompts and outputs stored, and can we set retention to zero?
3. Where are the servers located, and who are the subprocessors?
4. Is data encrypted in transit and at rest?
5. Does the vendor have a SOC 2 Type II report or ISO 27001 certification?
6. Will the vendor notify us of a breach, and how fast?
7. What happens to our data when the contract ends?

Read the terms of service yourself. A sales rep’s answer on a call isn’t a contract. And free or consumer versions of popular tools often have weaker terms than the business or enterprise versions.

Keep Identifying Details Out of Prompts

The easiest way to protect client data is to not share it. A lot of AI tasks don’t need real names or case facts to work.

Train your team to strip out or replace:

• Client and opposing party names
• Social Security, account, and case numbers
• Addresses, dates of birth, and phone numbers
• Medical, financial, and immigration details
• Names of children in family matters

Use placeholders like “Client A” or “[Spouse]” instead. For example, instead of pasting a full settlement email, ask the tool to “draft a polite reply declining a settlement offer and asking for a meeting next week.” You get the same draft without exposing anything.

Redaction isn’t perfect. A unique set of facts can still point to one person, even without a name. So the rule should be: if the task truly needs confidential details, use only an approved tool with strong contract terms, and get client consent where the rules call for it.

Write an AI Use Policy Your Staff Will Actually Follow

The Florida Bar opinion says firms should set policies for reasonable oversight of AI. That covers lawyers, paralegals, assistants, and anyone else who touches client files.

A useful policy is short. It should cover:

• Approved tools. List the exact tools staff may use. Keep the list small. Everything else is off limits for client work.
• Data rules. Spell out what can and can’t go into a prompt.
• Review rules. A lawyer must check every AI output before it goes to a client, a court, or opposing counsel.
• Personal accounts. Ban the use of personal AI accounts for firm work.
• Billing. Explain how AI time and costs are billed, in line with Rule 1.5.
• Reporting. Tell staff who to call if they think client data went somewhere it shouldn’t.

Then train people on it. A one-time email won’t stick. Short sessions with real examples work better, and so does a refresher when you add a new tool. Staff who want to understand how AI models handle data can find data science and AI learning resources that explain the basics in plain terms. People follow rules better when they understand the reason behind them.

Technical Controls That Back Up the Policy

Policies depend on people remembering them. Technical controls catch what people miss. You don’t need a big IT budget for most of these.

• Single sign-on and multi-factor authentication. Staff log in to approved AI tools with firm accounts, not personal emails.
• Role-based access. Only people working on a matter can upload or query its files.
• Audit logs. Pick tools that record who used them and what was shared. These logs matter if a court or the bar ever asks how AI was used.
• Zero data retention settings. Turn them on where the vendor offers them.
• Data loss prevention (DLP). DLP software can flag or block Social Security numbers and other sensitive data before they’re pasted into a website.
• Block unapproved sites. Your firewall or browser settings can stop access to AI tools that aren’t on the approved list.
• In-house or private deployments. Larger firms can run AI inside their own cloud environment so data never leaves their control.

Review these settings at least twice a year. Vendors change their terms and features often, and a setting that was safe last spring may not be today.

Talk to Clients About AI Before You Use It

Clients have a right to know how their information is handled. Under ABA Opinion 512, informed consent means a real conversation. That includes what kind of information the tool would see, who could access it, and what the risks are.

Your engagement letter is a good place to start. Add a short section that explains:

• Which types of AI tools the firm uses
• What those tools are used for, such as drafting or document review
• How the firm protects client data in those tools
• How the client can say no

Some practice areas need extra care. Family law is a clear example. A divorce or custody case can include financial records, medical history, and details about children. A fort lauderdale family lawyer handling that kind of matter has strong reasons to limit what goes into any outside AI tool and to explain those limits to the client up front.

If a client says no, respect it. Note the decision in the file and make sure everyone on the matter knows.

Have a Plan for When Something Goes Wrong

Mistakes will happen. What matters is how fast you respond. Your plan should say:

1. Who staff report to, right away
2. How to ask the vendor to delete the data
3. How to check whether the data spread further
4. When to tell the client, as your duty of candor requires
5. Whether Florida’s breach notice law or another state’s law applies

Write this plan before you need it.

The Bottom Line for Law Firms

AI can make legal work faster. But the duty to protect client data hasn’t changed. Pick vendors carefully, keep sensitive details out of prompts, set clear rules, and talk to clients openly. Firms that do these things can use AI and still keep the trust their clients place in them.