AI Recording Tools and Attorney–Client Privilege in the Age of AI

Key Takeaways

• Recording platforms that rely on AI can jeopardize attorney–client privilege if conversations are stored, shared, or used for model training by third-party vendors.
• Automated note-taking alters the confidentiality analysis: Transcription, summarization, and cloud-based processing introduce new exposure points even when recording is lawful and consensual.
• Accuracy in AI-generated records is not neutral: Summaries may flatten nuance or overstate legal conclusions, creating risk if lawyers or clients rely on them without careful review.
• Employers are hiring for skill combinations rather than titles: Data from the Wharton–Accenture report shows growing demand for hybrid legal and technical capabilities over traditional role-based credentials.
• The spread of AI is reshaping which legal skills carry value: Routine cognitive work is declining in importance, while judgment, regulatory expertise, domain knowledge, and client-facing skills are increasingly rewarded.

Final Thoughts:

AI tools can strengthen legal practice—but only if lawyers understand the boundaries that protect trust, confidentiality, and professional responsibility.
As recording technologies become more embedded in everyday interactions, lawyers will need to make deliberate, informed choices about when convenience aligns with their ethical obligations—and when it does not.

AI Recording Tools and Attorney–Client Privilege in the Age of AI

Introduction

Jen Leonard: Hi everyone, and welcome back to AI and the Future of Law, the podcast where we explore all of the exciting developments in the world of artificial intelligence and talk through what it might mean for lawyers and the legal profession. I’m your co-host, Jen Leonard, founder of Creative Lawyers, here live again in Philadelphia with Bridget McCormack, president and CEO of the American Arbitration Association. Hi, Bridge.

Bridget McCormack: It’s so cute that you call me Bridge. I love it. My dad used to call me Bridge, and my sister still does. I love that.

My excuse to be in Philadelphia. The ALI Council is meeting and working on Restatement drafts, so I did that all day. And now it’s fun to talk with you about AI and the future of law.

Jen Leonard: Very cool. I’m thrilled you’re here—Bridge in Philadelphia, the city of brotherly love and sisterly affection.

So we’re going to dive right in. We have three segments to our podcast: our AI Aha!’s, where we share something we’ve been using AI for; our What Just Happened, something in the broader landscape of AI development and what it might mean for lawyers; and then our main topic, which is explicitly legal.

And today we’re going to be talking about using AI recordings in your legal practice—something that I hadn’t really thought a lot about before. So I’m excited to talk about that with you and get your thoughts. But before we do, could you get us started with your AI Aha!?

AI Aha! Moment

Bridget McCormack: Well, inspired by all of the fab coding that everybody was doing, I decided it was time for me to start experimenting with vibe coding. I wanted to start in a very low-stakes context.

I think I’ve told you that my friends and I come up with new names for cocktails. We don’t really know what’s going to be in the cocktail—we just send the names to our little group chat. So things like Season of Hard Decisions and Boarding Group Six. I really wanted to make Ignatius J. Reilly. I don’t know if you’re a Confederacy of Dunces fan, but Ignatius J. Reilly is one of the most amazing characters in any novel ever. You really have to read A Confederacy of Dunces.

Anyway, I decided that now I could code an app where I give it the name of the cocktail and it gives me the ingredients—and tells me why.

So first I tried it in Claude Code, because you told me. Everyone was saying that should make it really easy. And the interface did make it seem like it was really easy. It produced something that looked like an app where I could put in the name and it was going to tell me the ingredients, but it didn’t work.

I asked it to try to figure out where the bugs were, and I thought I was doing what everybody said. I never got it to work.

So I went to ChatGPT. Their Codex tool is, by all accounts, similar—maybe not quite as good, based on what people were saying over the holidays—but I actually had better luck there. I gave it the name of a cocktail, and it gave me the ingredients.

I tried to write down some of my notes. For Season of Hard Decisions, the cocktail is designed to be a slow sipper—complex, slightly bitter, and strong enough to encourage contemplation. It balances the bite of the decision with the weight of the consequences.

It’s two ounces of rye whiskey, high proof—preferably 100 proof—half an ounce of amaro, something distinct like Cynar or Fernet Branca for an earthy, bitter edge, a quarter ounce of elderflower liqueur to represent the sweet relief of a choice, two dashes of bitters, one dash of orange bitters, and a garnish of a lemon twist, expressed and discarded, with a single Luxardo cherry at the bottom representing the reward.

And then it tells you the instructions and why each ingredient is important. It’s basically an amazing app, and I’m pretty sure I’m going to get rich. So I don’t know if I’ll be here for our next episode, because I probably made a hard decision. My cocktail app is obviously going to go viral.

Jen Leonard: Well, congratulations on your soon to be wild success as an app startup founder.

Bridget McCormack: What do you have for me?

Jen Leonard: Mine’s not nearly as entertaining, but I did want to talk about Google Sheets for a minute, because it’s kind of under the radar and it’s gotten so good. 

I’m not someone who really understands how to maximize Excel or Google Sheets. Google Sheets was already supposed to make Excel more accessible in the Google-y way they do things.

But now Google Sheets has Gemini integrated into it. You can open a Google Sheet, click on the Gemini icon, and tell it exactly what you want the spreadsheet to do. It will show you a sample spreadsheet with dropdowns.

You can say things like, “I don’t like column B—can you take that out?” or “I want this to be a dropdown menu instead,” or “Here’s something else you didn’t include that I really need.” And then it regenerates and does it again. And you can talk to it in natural language about formulas too.

My daughter is a Girl Scout. I love the Girl Scouts. I hate cookie season. You feel like you finally get good at something—and then all of a sudden it’s totally irrelevant.
You get to this point in the school year where you’ve done all the back-to-school things, you’re on all the platforms, you know how to do everything, you know everybody’s teacher. And then you get that dreaded email: it’s Girl Scout cookie time.

This is a whole project unto itself. But not anymore. I opened Google Sheets and said to Gemini, “My daughter’s a Girl Scout and I need a spreadsheet to keep us organized for cookie season. Can you help us?”

That’s all I said. It created a spreadsheet with the name of the person buying, the number of boxes, and it pre-populated all of the cookies for the season.

Here’s where I made adjustments. It had a column for payment—cash, check, credit, Venmo—and we don’t take checks. I just tell them we’re modern Girl Scouts. We use Venmo now.
So I asked it to eliminate checks, and it did. Then I noticed the cookie-type column wasn’t a dropdown, so I asked it to make that a dropdown so I could easily adjust it. It did that.
It didn’t initially have the formula to multiply the number of boxes by the price per box, which is $6—so I asked it to add that. Inflation has hit the Girl Scouts. Now the column at the end calculates totals automatically.

I’ve been using Google Sheets a lot more since then—for business purposes too. I asked it to help me create a leads and engagement spreadsheet. I didn’t even really know what people put in those. It helped me think through things like percentage likelihood of getting a client—75% for a strong lead, 25% for a weaker one—so you can start doing revenue projections. I assume most business people already know how to do this. I did not.

Bridget McCormack: I have two follow-up questions. You said it automatically filled in the cookies. Did it pull those because Gemini knows what cookies are on the menu this year?

Jen Leonard: I asked it to go to the Girl Scouts website and find the flavors for this year, and then populate them into the spreadsheet. So it did that.

Bridget McCormack: Second question: did you add a column for tariffs? Because you might want to put tariffs on those coconut ones

Jen Leonard: A 200% tariff. That’s a great column to add.

What Just Happened

Bridget McCormack: Well, it sounds like you’ve developed a whole lot of new skills, which brings us to our What Just Happened segment.

There’s this cool report, based on what might be the most data I’ve ever heard of anything being based on, from Accenture and Wharton together, on skills in the workforce. Tell us what we learned from this interesting report.

Jen Leonard: First, a shout-out again to our fabulous producer, Aaron Tran, who sent this our way. So this is a new report. It’s a research tool and benchmark created jointly by Wharton and Accenture. What they’re trying to do is analyze what’s happening in the labor market, not from a macro job-title perspective, but by really drilling down into the skills that matter.
This was an enormous analysis. They looked at over 150 million worker profiles and 100 million job postings. They were trying to understand which skills are oversupplied in the market versus undersupplied, which skills correlate with higher pay, and how AI is changing skill demand over time.

It’s a really interesting way to structure the analysis, and there were three main findings.
The first is a signaling gap between workers and employers—between what workers emphasize and what employers are actually looking for. They’re kind of talking past one another.
Workers emphasize things like leadership, communication, teamwork, and problem-solving—generalist skills. And in fairness to workers, last year LinkedIn put out a list of the most in-demand skills, which track pretty closely to this. So applicants are following the trends and trying to respond to them.

But it turns out that what employers actually reward are hard technical skills: technical depth, understanding scientific methods, analytical precision, digital execution, operational expertise, and domain-specific knowledge.

So the skills everybody is putting on their resumes are in enormous surplus, while the skills employers are looking for are in persistent deficit. In a lot of cases, resumes have just become meaningless noise. I thought that was really interesting.

Bridget McCormack: And that seems relevant not only for whether you can get a job, but also for compensation. According to the study, you’re paid more if you fit into one of these deep domain skill areas—not necessarily aligned with a job title, but aligned with the skills themselves.

Jen Leonard: The riches are in the niches. And it’s not that surprising when you step back and think about it. Even in law firms, what we hear is, “We need people with this discrete set of skills.”

Bridget McCormack: The AAA has over 800 employees, and like everyone else, we’re operating in a rapidly changing market. Some of the things we’re hiring for, we don’t even have names for. I think I’ve talked to you about one of the roles we’ve been hiring for lately—I’ve been calling it a “legal technical centaur.”

They need legal skills—they’ve been to law school (not just vibe law school)—but also technical skills, because they’re working at the governance intersection of both.

So we don’t really know what to call the job. We don’t have a name for it.

Jen Leonard: I think you should call it a “legal technical centaur”.

Bridget McCormack: I think you’re right. But my point is that it’s really baskets of skills that matter more and more to employers.

I don’t need somebody who can only do law anymore. And I don’t need somebody who doesn’t know anything about law at that governance intersection either. I need somebody who can operate in the center.

Jen Leonard: It’s so true. When you think about traditional job titles—marketing specialist, law firm associate—I see this even in our tiny little business. And you see it in a large organization too. We’re looking for very specific skill sets, and sometimes we can’t even figure out how to describe what we’re looking for.

Which leads to the second finding—that skills have price tags, and they’re role-specific. The same skill can increase your salary in one role and decrease it in another.

The example from the report comes from life sciences. For a validation lead, which is a technical role, saying you have strategic analysis capabilities actually correlates with lower salaries. Employers aren’t looking for broad skills there.

But for a sales representative in the same field, that same strategic analysis skill correlates with salaries that are nearly $8,000 higher.

So there’s no universal high-value skill anymore. You really have to empathize with the employer and ask: is this a value-add for what they’re looking for, or does it actually make me look more generic in the market?

Bridget McCormack: I still stand by those general skills, though—growth mindset, bias in favor of curiosity. Once you’re inside an organization, those things really matter.

Jen Leonard: They’re just harder to suss out. Maybe one takeaway for applicants is to emphasize discrete, technical skills to get in the door, and then demonstrate those broader skills in the interview process.

The third takeaway is that AI itself is redistributing value across skills. Things that were valuable for most of our lifetime are becoming less valuable, and other skills are coming to the fore.

The report tracked how demand for skills has changed since the ChatGPT moment in late 2022. There’s declining demand for writing, simple analysis, and structured, repeatable cognitive work—which makes sense, because AI does those things very well.

At the same time, there’s rising or stable demand for regulatory and compliance skills, operations management, judgment and coordination, domain expertise, and relationship-driven capabilities like sales and communication.

AI is shifting which skills matter. The skills that require context, judgment, and accountability are becoming the most valuable, because those are the bottlenecks where humans really have to be involved.

Bridget McCormack: I mean, it all makes sense, and I think it’s pretty valuable information. It’s definitely a report worth—if you’re not going to read it—putting into NotebookLLM and listening to the podcast about it. Especially if you’re thinking about a job change or getting out of law school. 

Jen Leonard: Or you could take this podcast, put it into NotebookLM, and turn it into a different podcast. Oh my gosh. It’s like that movie with Michael Keaton where they keep making copies of Michael Keaton and they keep degrading over time. Multiplicity, right?

So what are the recommendations for the different groups? For employers, the report says they should stop hiring by job title—unless it involves “centaurs,” which I think speak for themselves—and instead decompose roles into skills and align pay with the actual skill economics, rather than what that job traditionally pays in the market.

For workers, the recommendation is to think about your career more as a skill portfolio—which I really like—and to use AI to build technical depth faster, which is what we’re constantly doing.

And for educators, the recommendation is to shift curricula toward specialized, job-ready skills and to teach students how to signal their skills with specificity. There’s a lot there for thinking differently about the job search. It’s a hard time to be looking for a job.

Bridget McCormack: And for employers trying to think about their org charts. Yesterday’s org chart is kind of useless. That’s probably a good thing for workplace culture, even though it can feel scary if you’ve spent a long time in a system where you move from this position to that position.

It can feel destabilizing at first, but I actually think it’s exciting to imagine org charts built around sets of skills and clusters of people with those skills working together on different projects. There’s a lot of room to rethink org charts in ways that are genuinely good for organizations.

Jen Leonard: I think it’s so exciting. I remember working in career services with students about ten years ago, when we talked all the time about resume red flags—like not moving around too much, or not taking certain roles because it might signal you’re a flight risk.

Looking back, that world feels like a recipe for anxiety and staying stuck, whether you like it or not. It’s liberating to think about a much more fluid world, where you can make your work more meaningful over time without those constraints. I think it’s a hopeful story, and it’s a really interesting report.

And that brings us to our main topic, which also relates to skill shifting—thinking about new technologies and how they affect the way we’ve traditionally thought about legal practice. We’re going to talk about AI recording in legal practice, some of the benefits of using AI recording tools, and some of the things lawyers really need to think about before they do.
So, Bridget, maybe you could walk us through this issue at a high level before we dig in.

Main Topic: AI Recording in Legal Practice

Bridget McCormack: Yeah. Lawyers have been using recording devices with clients and witnesses—in informal settings—for a long time. And they’ve always had to think about the legal, regulatory, and ethical issues that govern that practice.

People are often familiar with state laws around recording. In some states, including Pennsylvania, where we are right now, there’s criminal liability if you record someone without their knowledge. That’s not just true for lawyers; it’s true for anyone. You can’t record someone without their consent. But that means, it’s also true for lawyers. In some states, it’s perfectly acceptable and not a crime.

In Pennsylvania, it’s a felony. And that’s actually not uncommon. There are a number of states where that’s the case.

New York is different. There’s no criminal prohibition on recording someone without their consent. But that doesn’t mean there aren’t other regulatory or ethical issues. For example, you could waive privilege in certain circumstances. In some states, there are professional rules of conduct that prohibit deceptive practices. We’ll talk a bit about a New York ethics opinion on that issue.

AI changes this a little bit, because it’s not like turning on a tape recorder in your pocket in a state where that’s permitted. If you’re using one of the many AI recording devices on the market, you also have to think about what that device is doing with the recording.

These tools are pretty amazing. They’ve gotten really good at listening to meetings—even panels—and taking detailed notes, then organizing them for you. But the question is what those recording devices are doing with that information. And does what they’re doing with it, waive the attorney–client privilege.

If you’re recording a witness or a panel, privilege isn’t really the issue. But if you’re recording a conversation with your client, you need to know what the vendor is doing with that information—whether it’s being shared in the cloud, used in training data, or stored in some other way.

And does that violate the attorney–client privilege? Because the traditional rule is that if a third party is privy to that conversation, it’s no longer privileged.

That’s really significant, because one of the most fundamental guarantees lawyers make to their clients is that they can speak confidentially, and that what they say is protected. That guarantee is foundational to why the relationship works.

So there’s a gating issue right now if you’re using any AI recording device—even if you’re just on a Zoom meeting and using Zoom’s recording function, or Teams, or other video platforms. All of these tools have recording settings, and you need to understand what they’re doing with the data.

I think Zoom and Teams now have settings—especially in enterprise versions—that address this, because it came up pretty early. But there are also a lot of startups on the market. Otter AI, for example, which many people use, has been subject to a class action lawsuit related to how data was used for training.

So for anyone, it’s worth thinking through these issues. But for lawyers, who have ethical obligations, it’s especially important to understand the tools you’re using, whether there’s an issue with attorney–client privilege, and whether you have the right version of the product and the right settings in place.

There’s also a New York ethics opinion that raises the stakes around deceptive practices. New York Bar Formal Opinion 2025-6, which was issued right before the end of the year, concludes that Rule 8.4’s prohibition on deception is violated by secret recording—even though secret recording isn’t illegal in New York. But it’s unethical for a lawyer.

So again, this is something lawyers should think through carefully with respect to conversations with clients, the devices and platforms they’re using, and the potential consequences. These tools can be incredibly useful for lawyers, but there’s a separate set of questions lawyers need to ask.

Did I miss anything important?

Jen Leonard: No, not at all. That was such a great overview. One thing that really caught my eye—and that I hadn’t fully thought through before—is the implication for the duty of loyalty that the opinion draws out.

Clients might speak differently if they know they’re being recorded. And to do your job well as a lawyer, you need to know everything from your client.

That feels like a hard thing to suss out—thinking through when you’d use recording and when you wouldn’t, even if your client is okay with it, and even if you’ve resolved the privilege issues.

Bridget McCormack: If I were being recorded, even just as a human, I think I’d be very careful about choosing my words. I might hold things back. And that doesn’t seem helpful in an attorney–client relationship.

I’m also not sure I fully understand why this is framed as duty of loyalty rather than competence. To me, it feels like competence—you want to make sure the relationship is one where you can do your best work, and that depends on getting the best information.

Jen Leonard: I also wonder whether this will become a generational issue over time. I’m Gen X. I wasn’t raised with people taking pictures of me everywhere and posting them online.
When I worked at a law school, we’d have student events where there was a notice saying, “You’re being photographed.” I found it creepy. I’d try to take photos quietly, and the students would say, “Do you want a better shot? I can give you my good side.” They were completely used to it.

So I wonder whether, as these tools become more embedded in our surroundings, the concern that clients might change how they speak because they’re being recorded will fade—or maybe not.
And there’s the other side of the coin, which is clients recording conversations. 

Bridget McCormack: There are clients who want to record the conversation. I actually have a lot of empathy for that. Clients want to remember what they’ve been told.

We recently met with a lawyer to do wills, and the minute I left the office, I couldn’t remember anything he’d said. My husband asked me questions, and I wasn’t sure. If I’d had a transcript, it would have been helpful—and I wouldn’t have had to email him again.

But that raises the same problems. What device is the client using? Is the information being sent to the cloud? Are we putting privilege at risk as a result? So it really does go both ways.
Is it becoming good practice for lawyers to ask clients directly—“Just so you know, I’m not recording this conversation, and I want to make sure you’re not recording it either”? Is that a conversation lawyers need to have at the beginning of a relationship?

Jen Leonard: The opinion suggests including this in the retainer or engagement letter, which is probably best practice. Of course, nobody reads those. I’d probably drop it into ChatGPT and ask for a summary.

But it’s worth asking whether this is happening in other professions too. People bring tools like Otter AI to doctor appointments because they don’t remember what doctors say. I have empathy for that.

I also have empathy for both sides of the lawyer–client relationship. Lawyers have obligations, but these tools can also make you better at lawyering. You can be fully present in the conversation instead of trying to capture every detail.

One risk I’ve heard discussed is AI misinterpreting nuance. A lawyer says, “I think we have a strong case,” and the AI-generated notes turn that into, “We’re going to win.” That loss of nuance can be a real problem. Over time, the systems may get better, but right now it’s something to be mindful of.

Bridget McCormack: If you’re a client who’s stressed and trying to absorb information, you might decide that imperfect notes are still better than walking out with no memory at all. So there’s a trade-off.

Jen Leonard: This feels similar to how organizations think about AI generally—using a risk-based framework. An intake conversation that’s basic and informational might be a safe use case. A capital murder case, where you’re discussing a prosecution’s offer, is probably not something you want to record.

So the highest-stakes matters are likely where you either wouldn’t use these tools or would be extremely cautious. And over time, we’ll probably see new fact patterns and more guidance emerge.

This is a new issue for me, and one I’ll be interested to keep following—both how lawyers use these tools and how they sometimes get into trouble with them, just like we saw with early ChatGPT hallucination issues.

Bridget McCormack: Definitely.

Jen Leonard: Thank you for walking us through this, Bridge. And thank you to everyone for joining us. We look forward to seeing you on the next episode of AI and the Future of Law. And until then—take care, and don’t record anybody without their consent.