MyShingle Ethics Opinion 2026-2

The One With the Ethics of AI Notetakers for Attorney-Client Conversations.

Issued July 11, 2026 · MyShingle.com

Background

Advances in AI-powered transcription combined with the post-pandemic surge in online meetings have made it easier than ever for lawyers to record and document attorney-client communications. The appeal is obvious. Recording a client meeting lets the lawyer focus on the client instead of a legal pad, and a transcript catches details that even a good notetaker misses, sparing the office costly and embarrassing follow-up calls. Clients benefit too. A transcript memorializes the advice given, serves as a reference the client can consult later (how the divorce process works, what the incorporation steps are), and fosters transparency and trust. Transcripts also protect the firm, cutting down on after-the-fact “I said, you said” disputes and letting lawyers monitor whether support staff are dispensing legal advice, a concern the ABA addressed in Formal Opinion 506 (2023).

But the same features that make AI notetakers attractive raise questions about consent, confidentiality, and competence. This opinion addresses the question: May a lawyer use AI tools to record, transcribe, and summarize conversations with clients consistent with the Rules of Professional Conduct, and if so, under what conditions?

A lawyer may use AI notetaking tools to record, transcribe, and summarize client conversations but only with disclosure to the client and, in most circumstances, the client’s consent. Undisclosed recording is deceptive under Rule 8.4 even in one-party-consent states, and at least one regulator has now said so expressly. Before turning the tool on, the lawyer must understand how it handles data, such as where recordings are stored, whether the vendor trains on them, and whether deleted content is actually deleted.  Lawyers must also steer clear of consumer-grade tools that train on client data. Rule 1.6 requires prudent use even of a vetted tool: keep highly sensitive details out of recordings, download transcripts to firm storage, and delete them from the vendor platform promptly. Lawyers must review AI-generated transcripts and summaries for accuracy before relying on them, and must comply with state biometric privacy laws now on the books in Illinois, Texas, Washington, and several other jurisdictions that may require informed written consent before a client’s voice is recorded.

Rules discussed: ABA Model Rules 1.1, 1.4, 1.6, 8.4

Introduction

Guidance on recording clients long predates AI. ABA Formal Opinion 01-422 (2001) concluded that a lawyer may not record conversations in violation of law in an all-party-consent jurisdiction, nor falsely deny that a recording is being made  but the ABA committee divided on whether recording a client without the client’s knowledge is unethical, calling it merely “inadvisable.” State authorities split the same way: at least eighteen states have concluded or implied that recording client conversations without permission is not unethical, while at least nine have held or implied that it is unethical even where legal. (Justia maintains a 50-state survey of state recording-consent laws.) One commentator argues that secret client recordings are inherently deceitful and violate Rule 8.4 outright. J. Bries, Legal Ethics of Secret Client Recordings, 24 Geo. J. Legal Ethics 55 (2020).

Although only a small percentage of attorneys considered recording clients back in the day of the cassette tape recording, transcription is gaining traction in the AI era. AI notetakers are much easier to use than predecessor hardware and boast added capabilities to join meetings automatically, transcribe verbatim, generate summaries, store content on vendor clouds, and — depending on the terms of service — to use recordings to train models. The first regulator to confront these tools directly is the New York City Bar, whose Formal Opinion 2025-6 concludes that lawyers must notify clients and obtain consent before recording with AI-enabled tools, and that undisclosed recording is deceptive under Rule 8.4 and inconsistent with the duty of loyalty even if only a summary is ultimately retained because clients speak differently when a verbatim record is being made.

This opinion synthesizes that guidance for the solo and small firm lawyer, who evaluates vendors personally, signs the engagement letter personally, and bears the malpractice risk personally.

Discussion

Competence (Rule 1.1)

Comment 8 to Rule 1.1 requires lawyers to understand “the benefits and risks associated with relevant technology,” a duty roughly forty states have adopted. For AI notetakers, competence means the lawyer can answer basic questions about the tool before using it with clients: Where are recordings stored? Who can access them? Does the vendor use recordings or transcripts to train its models? What are the default recording settings? Does deletion actually remove the data? NYC Formal Opinion 2025-6 makes these inquiries part of the competence duty itself.

Answering them requires reading the AI notetaking platform terms of service.  Although AI can produce a first-pass summary of the terms, the lawyer remains responsible for the conclusions. AI notetaking tools built for lawyers generally include adequate safeguards, but “built for lawyers” is marketing, not a warranty; lawyers must verify the salesteam’s claims. Reject any platform that trains on the entirety of your recordings, retains content after you delete it, or fails to take reasonable steps to protect confidentiality. Where the terms leave doubt, ask the vendor directly, and consider private-cloud options where offered. MyShingle’s review of the Fireflies.AI terms of service and its data-protection guide illustrates the exercise: the terms commit to reasonable privacy and security measures and permit complete deletion, with data used only in limited, anonymized form to improve transcription. This satisfies confidentiality requirements for most routine client communications, though mileage varies with the sensitivity of the matter.

Competence also means reviewing the output. AI transcripts and summaries contain errors, and a verbatim record can transform informal, “off the cuff” advice into something that reads like a considered opinion. Review transcripts and summaries for accuracy before filing them or sharing them with the client.

Communication and Client Consent (Rules 1.4 and 8.4)

The better rule — and now the express rule in New York City — is that lawyers must tell clients the recorder is running and obtain consent before it starts. NYC Formal Opinion 2025-6 holds that undisclosed AI recording is deceptive under Rule 8.4 and violates the duty of loyalty regardless of one-party-consent law, and regardless of whether the lawyer keeps only a summary. That reasoning will likely travel: the client is owed candor from her own lawyer above all.

Disclosure is also simply better practice. Secret recording, even where legal, destroys the very transparency and trust that a shared transcript is supposed to build. Don’t just disclose.  Instead, explain why you record, offer the client the transcript, and give the client the chance to opt out. Disclosure belongs in the engagement agreement, which as discussed below in connection with state biometric privacy laws is also the natural place to capture written consent. A sample provision:

Use of AI Meeting Tools. To give you our full attention and an accurate record of our advice, we use secure AI tools to record, transcribe, and summarize certain client meetings and calls. By signing this agreement, you consent to that recording, including the processing of your voice, which some state laws treat as biometric data. You may opt out at any time, for any or all conversations, with no effect on our representation. We protect these records as confidential client communications, use only vetted tools that do not train AI on client data, avoid recording highly sensitive details, and review all transcripts and summaries for accuracy; until we adopt one, it is a draft, not our legal advice. We ask that you not record with your own tools, because consumer apps can risk waiving attorney-client privilege.  We are always happy to share our transcript with you instead.

The consent conversation should run in both directions. When clients record with their own consumer apps, the lawyer loses control of the output since a recording can be edited, and insecure tools can waive privilege. In New York, lawyers may include an engagement provision expressly barring clients from recording attorney-client conversations. N.Y. State Bar Ethics Op. 1270. Even without going that far, explain the risks of DIY recording and offer your transcript as the substitute; the Illinois Supreme Court Commission on Professionalism offers additional tips on handling client AI notetakers.

Confidentiality (Rule 1.6)

Comment 18 to Rule 1.6 requires “reasonable efforts” rather than Fort Knox level protection to prevent unauthorized disclosure.  Lawyers should weigh the sensitivity of the information against the cost and difficulty of added safeguards. Applied here, reasonable efforts means layering prudent habits on top of a vetted tool: avoid recording trade secrets, or other highly sensitive material; download transcripts regularly to your own storage; and delete recordings from the vendor platform as soon as practicable.

Multi-party meetings deserve special caution. When settlement or negotiation participants each bring their own AI notetaker, every party must consent to every recorder — and as host, you may be on the hook when consent is missing, with no control over how the recordings are edited or disseminated later. Block uninvited AI scribes, announce in advance that notetakers are not permitted, and if a recording would be useful, designate one recorder that shares the output with all participants.

Compliance with State Biometric Privacy Laws

Ethics rules are not the only constraint — lawyers must also comply with federal, state, and local law. As the Federal Trade Commission has explained, biometric information includes data derived from depictions, images, or recordings of a person “to the extent that it would be reasonably possible to identify the person” from that data — a definition that may reach voice recordings. As of July 2026, several states, including Illinois, Texas, and Washington, and even local governments such as Erie County, New York, have adopted biometric privacy laws requiring informed, written consent before biometric information is collected.

Whether every statute reaches voice recordings is not yet settled, but the safe course is cheap: obtain written consent before the recording begins or, better, in the engagement agreement itself, using a provision like the one above. And because this body of law is expanding quickly, lawyers should track biometric and AI privacy legislation in the jurisdictions where they practice as part of the ongoing Rule 1.1 duty to keep abreast of changes in the law.

Conclusion

A lawyer may use AI notetakers with clients, and often should in order to enjoy the gains in accuracy, attention, and client transparency. To satisfy ethics obligations, lawyers should disclose and obtain consent, preferably in writing in the engagement agreement; vet the tool’s data handling before the first recording; use even a vetted tool prudently; review every transcript and summary before relying on it; block uninvited notetakers in multi-party meetings; and track the state biometric privacy laws that increasingly govern voice recording. Undisclosed recording of a client fails Rule 8.4 under the better view and fails the client under any view.

This area is moving fast. NYC Formal Opinion 2025-6 is the first regulator’s word on AI recording, not the last, and biometric privacy statutes are multiplying. Lawyers may expect the consent-and-disclosure baseline described here to remain in effect moving forward

MyShingle Ethics Opinions are advisory commentary by Carolyn Elefant to provide guidance to attorneys but similar to Ethics Hotline guidance is not binding and does not constitute legal advice. These opinions may be cited by researchers, ethics committees and courts. Lawyers should consult the rules and opinions of their own jurisdiction.

MyShingle Ethics Opinion 2026-1

The One About Passing AI Costs Through to Clients

Issued July 10, 2026 · MyShingle.com

Background

On July 12, 2026, Anthropic is expected to shift access to Claude Fable 5 from subscription bundles to metered, pay-per-token usage available only through usage credits or API calls, so each serious use will generate a per-token charge. Pricing trackers put Fable 5 at about $10 per million input tokens and $50 per million output tokens, making it Anthropic’s most expensive generally available model and roughly twice the listed price of Opus 4.8.

Fable is just one of many price increases as law firms accelerate their adoption of AI. Agentic AI is also pricier than the chat bots of two years ago.  Last month, Artificial Lawyer predicted that token costs for large law firms could reach millions of dollars, while many businesses claim that their AI bills now exceed the cost of the workers AI was intended to replace. The market is already responding: Legora announced consumption-based pricing for its agentic models, presumably to shift responsibility for burgeoning AI costs from the platform to users. Commenters are asking here and here whether lawyers may appropriately pass the cost of AI on to clients.

That is the question presented in this inaugural MyShingle Ethics Opinion: May lawyers pass the costs of artificial intelligence, from subscriptions and platform licenses to metered per-token usage charges, through to clients, and should they?

AI costs reach a law practice in two forms, and the answer differs for each. Subscriptions, platform licenses, and tools that equip the practice generally are overhead: a lawyer may not bill them to clients or pro-rate them among clients, any more than the lawyer could bill for the research subscription or the case management system. Metered charges are different. A lawyer may charge a client for metered AI usage, but only when the charges are attributable to the client’s specific matter, billed at actual cost without markup, disclosed in advance and preferably in writing, and not duplicative of costs the lawyer already recovers as overhead. A lawyer who cannot isolate the actual cost for a particular matter may not charge the client at all. But the ability to trace a charge to a client’s matter does not by itself justify billing it.  It is our position that as a matter of policy, ordinary AI usage should be absorbed as overhead even when the platform attributes every charge to a particular client. Pass-through billing is appropriate in only two situations: extraordinary AI consumption attributable to a single matter and engagements in which the client expressly directs the lawyer to use a particular tool. A lawyer whose ordinary AI costs rise should raise rates or adopt flat fees rather than itemize tokens.

Rules discussed: Model Rule 1.5

Discussion

Costs in a law practice fall into two categories, and the classification controls whether they may be billed to clients.  ABA Formal Op. 93-379 (1993)  draws a line between overhead costs and client expenses.  Overhead is the cost of operating a law office and includes costs like rent and reception services, maintaining a library, malpractice insurance, purchasing utilities and the like.   Absent advance disclosure, overhead costs are not appropriately passed on to clients because (1) the client should reasonably expect that these costs would be subsumed within the charges for professional services” and (2) the benefits associated with these costs inure to the firm.  By contrast, client expenses are those costs incurred for a particular matter, such as filing fees, court reporters, and in-house services like photocopying. These may be billed to clients but only at “the direct cost associated with the service” plus “a reasonable allocation of overhead expenses directly associated” with providing it, and never with a surcharge absent client agreement. See ABA Formal Op. 93-379 (1993).  

The Westlaw Era: How the Profession Classified the Last Metered Technology

Technology and new business models can change how costs are characterized.  Consider the treatment of legal research tools which has evolved over time.  When lawyers conducted research in firm law libraries stocked with hard copy digests and reporters, they never passed the cost of books or librarians on to clients because these costs were considered overhead.  When   Lexis and Westlaw arrived in the mid-1970s, they charged by the search and by the minute of connection time or search units.

Because each Wexis search was attributable to a specific matter, the profession classified metered research as a client expense. Philadelphia Bar Op. 87-23 (1987) rejected that computerized legal research was overhead, but determined that the cost could be included as part of contingency fee recovery with advance disclosure to clients and an option to opt out.  And  two decades later, the prevailing view of federal circuits in fee recovery cases is that the cost of computerized legal research, if documented, was appropriately included in fee awards. See In re UnitedHealth Grp. Inc. S’holder Derivative Litig., 631 F.3d 913, 919 (8th Cir. 2011).

Then the vendors changed their pricing, and the classification flipped. In the late 1990s, Lexis and Westlaw began offering fixed-fee “commitment pricing” contracts, and a flat subscription has no per-matter cost. Firms that kept billing clients at retail per-search rates against flat contracts were no longer recovering an expense; they were marking up overhead. The practice persisted long enough to become a litigation risk: a 2021 lawsuit accused a large firm of billing more than $100,000 in per-search charges against its flat-rate LexisNexis contract.  Soon after, recovery of online research charges among large firms declined year over year beginning in 2005, 43 percent of surveyed firms reported absorbing more research cost by 2012 as clients balked or refused outright to pay, and computerized research settled where it remains today: “a cost of doing business and not a client recoverable cost.”

The AI Opinions

Every ethics authority to address AI costs has applied FO 93-379’s classification, and a consensus has formed around three rules. First, AI tools that equip the practice generally are overhead. ABA Formal Op. 512 (2024) gives the example of an AI grammar checker embedded in word processing software, whose cost “should be considered to be overhead”; the State Bar of Michigan’s AI FAQs extend the point to subscriptions and one-time tool purchases, which the firm absorbs absent advance disclosure; and Kentucky Op. E-457 (2024) adds that the cost of learning and keeping current with AI is overhead in all events. Second, per-use charges attributable to a specific matter are billable expenses at actual cost only. Opinion 512’s example is a third-party provider’s per-use charge for reviewing thousands of contracts for one client, which a lawyer may “ordinarily” bill; Texas Op. 705 (2025) permits recovery of per-use AI charges “in much the same way some clients agree to pay for the use of traditional online research tools like Westlaw and LexisNexis,” capped at the amount actually paid to the provider; and New York City Bar Op. 2024-5 (2024) routes AI costs through FO 93-379’s disbursement rules. Third, attribution must be real and disclosure must come first. Florida Op. 24-1 (2024) holds that a lawyer who cannot determine the actual cost associated with a particular matter may not prorate periodic AI charges and “should account for those charges as overhead”; New Mexico Op. 2024-004 requires the lawyer to disclose the actual cost and the intent to charge it, in writing, before the representation; D.C. Bar Op. 388 (2024) notes that an out-of-pocket AI cost “is an expense that should be communicated to the client,” and Opinion 512 forbids charges that duplicate amounts recovered from “this or other clients” and requires provider discounts to be passed along.

Two authorities depart from the actual-cost consensus. North Carolina 2024 FEO 1 permits AI cost recovery through either matter-specific charges or “a general administrative fee” covering technology expenses, with client consent, preferably written. The New York State Bar Association’s AI Task Force report (2024) would allow a surcharge above actual cost if clearly stated in the engagement letter and the total charge remains reasonable.

Classifying AI Costs

Applying that authority, AI costs are recoverable as follows. Subscriptions, platform licenses, and one-time tool purchases are overhead. They equip the practice as a whole, no method attributes their cost to one matter rather than another, and every authority to consider them so holds. They may not be billed, prorated, or recovered in any form other than the fee.

Metered usage is an expense only when (1) the charge would not exist but for a specific client’s matter and (2) the lawyer can measure its actual cost. So for example, under this test, firm-wide agents that monitor files and flags work to be done would fail the first prong regardless of token consumption, because it serves the practice generally.  Where both conditions are met, the charge may be billed at actual cost, without markup, without duplicating amounts recovered from any client, with provider discounts passed along, and with the charge and its basis disclosed in advance and in writing.  So the cost of usage tokens incurred by an agent that prepares and reviews discovery in connection with a specific client file could be passed through to the client.

Permissible, however, does not mean advisable. Just because firms can ethically recover metered charges doesn’t mean that they should.  Recall, FO 93-379 explained that recovery of overhead is not appropriate because clients expect overhead costs to be subsumed in hourly rates or the cost of legal services.  Likewise, in 2026, clients expect professional service providers to incorporate AI seamlessly throughout all aspects of practice.  Just as  clients wouldn’t expect a hospital to include a line-item charge for a robotic arm used in surgery, so too, they are increasingly less likely to expect a separate charge for AI research.  Moreover, token line item charges on an invoice smack of nickel and diming, and invite disputes over token counts. 

Even as token charges increase, law firms have options.  They can raise hourly rates, or charge a flat fee for service that covers all charges. 

Finally, there are two narrow situations that might justify pass-through. The first is extraordinary consumption attributable to a single matter. Because overhead is spread across every client through the fee, a five-figure AI bill for a single client absorbed as overhead forces every other client to pay for one client’s matter; the client whose matter created the cost should bear it. The second is the client-directed tool: when the client instructs the lawyer to run work through a specified platform, the charges are a client-directed expense, and may be billed at actual cost and documented in the engagement agreement.

Best Practices

Rather than run up hefty AI bills and pass them off to clients, the better discipline is to absorb AI costs and manage them prudently, applying these best practices:

• Match the model to the task. Reserve premium models like Fable for genuinely hard matters; use cheaper or bundled tools for routine emails, quick research, and form drafting.
• Reduce unnecessary tokens. Narrow prompts and chunk documents. Run a 50-page contract in sections and ask targeted questions rather than “analyze everything about this matter.”
• Turn on cost controls. Spend caps, per-matter budgets, and usage dashboards reveal which matters and users drive spend before the invoice does.
• Put guardrails on agents. Limit how many times an agent can re-run a chain, cache repeated context instead of re-sending full files, and avoid routing every step to the most expensive model by default.
• Audit the math periodically. Compare dollars spent on AI to hours saved and revenue generated, and adjust the mix of models and tools when the math stops working, both for your margins and for fees that remain reasonable to clients.

Sample Disclosure:  If you plan to pass on token consumption costs, here’s some sample disclosure language for clients:

Where the Firm incurs metered, usage-based AI charges directly attributable to your matter, those charges will be billed to you as a case expense at the Firm’s actual cost, without markup, and with any provider discounts passed on to you. These charges will be itemized on your invoice. If the Firm cannot determine the actual cost attributable to your matter, no charge will be made. By signing this agreement, you consent to this billing arrangement.

Conclusion

Are AI costs overhead or client expenses? Subscriptions and platform licenses are overhead and may never be billed. Metered charges attributable to a single matter may be billed at actual cost with advance written disclosure, but the view of this Opinion is that firms should absorb all AI-related costs as overhead subject to two narrow situations only: extraordinary single-matter consumption that functions like an outside vendor charge, and tools the client directs the lawyer to use.

MyShingle Ethics Opinions are advisory commentary by Carolyn Elefant to provide guidance to attorneys but similar to Ethics Hotline guidance is not binding and does not constitute legal advice. These opinions may be cited by researchers, ethics committees and courts. Lawyers should consult the rules and opinions of their own jurisdiction.

ATTORNEY-CLIENT COMMUNICATIONS IN THE AGE OF AI

With significant advances in the effectiveness of artificial-intelligence (AI) powered transcription combined with the post-pandemic rise in online video meetings, it’s now easier and more convenient than ever to record and document attorney-client communications with AI.

But should lawyers record and transcribe communications with AI?  Some lawyers are concerned about confidentiality and privacy or fear that a client might regard a recording as suspect.  This comprehensive post addresses the following questions:

• What are the benefits of incorporating AI-powered transcription into workflow?
• What ethical and legal considerations are implicated by recording conversations with clients?
• Do the Terms of Service for AI-recording platforms provide adequate protection for confidentiality?
• What best practices should lawyers employ when using AI to transcribe client communications?

1.  Transcribing Attorney-Client Communications Is A No-Brainer

Lawyers have always had ways to document client conversations through handwritten notes or staff-prepared summaries.  But where AI changes the game is by making documentation near-instantaneous and comprehensible – which was hit or miss when the task was left to humans. 

From a practice perspective, attorney use of AI to transcribe an initial consultation or trial prep meeting is really a no-brainer.  Foremost, recording a client meeting enables lawyers to focus on the client, without the distraction of taking notes. And a transcript will always catch details that a notetaker will miss, sparing your office from costly and embarrassing repeated follow up with  your client. 

Keep in mind that clients benefit from access to a recorded transcript also.  A recorded transcript fosters transparency and breeds trust with clients by showing your willingness to permanently memorialize your advice.  AI-transcripts also serve an educational function.

 If, for example, you explain how the divorce or incorporation process works during your initial client meeting, your client can consult the transcript to refer back to your instructions as you explained them.  And you can transcribe videos you may have  prepared on various processes for clients who prefer written instructions.  

Written transcriptions can inoculate your firm against ethics and malpractice claims.  A transcription of intake calls and substantive client communications cut down on after-the-fact ‘I said, you said” scenarios where the client recalls the conversation one way and you remember it another. Likewise, recording conversations between support staff and clients lets you keep tabs on whether legal advice has been dispensed by non-lawyers – an activity on which the ABA just issued restrictive guidance in  ABA Formal Opinion 506 on Responsibilities in Working With Non-Legal Assistants (June 2023).

2.  What Ethical and Legal Considerations Are Implicated by Recording Client Communications?

Lack  of Consent

The most significant legal and ethics problems arise out of non-consensual or surreptitious recordings of client communications.  ABA Formal Opinion 01-422 takes the position that “a lawyer may not record conversations in violation of the law in a jurisdiction that forbids such conduct without the consent of all parties, nor falsely represent that a conversation is not being recorded.”  However, the Opinion is “divided as to whether a lawyer may record a client-lawyer conversation without the knowledge of the client, but notes that it is inadvisable to do so.”

States’ views are mixed.  According to an article published in the University of Illinois Chicago Law Review, at least eighteen states have concluded or implied that recording client conversations without permission is not unethical  But advisory opinions in at least nine other states have held, or implied, that even if legal, it is unethical to record conversations without consent.  (For those curious, here’s a 50-state survey courtesy of Justia, current as of October 2022 summarizing state recording consent laws)

Finally, one legal scholar argues that secret client recordings are inherently deceitful, and therefore violate Model Rule 8.4.  See J. Bries, Legal  Ethics of Secret Client Recordings, 24 Georgetown J. Legal Ethics 55 (2020).

From a business perspective, secretly recording clients, even if legal in one-party consent jurisdictions, completely undermines the benefits of transparency and trust that flow from providing clients with a written record of attorney-client communications.  For that reason, full disclosure of client recordings, including an explanation of the reason for doing so is recommended as a best practice below for attorneys using AI to record and transcribe client conversations.

Confidentiality and Privacy

Recordings can also jeopardize client confidentiality if the information collected by the platform is used for training or made widely accessible. Nut regulators don’t mandate impenetrable Fort Knox level protection for every client communication. Comment 18 to ABA Model Rule 1.6 imposes a “reasonable efforts” standard to guard against disclosure, considering factors such as “sensitivity of the information” and cost and need for added safeguards.  Thus, even when using a reasonably secure recording product, lawyers should adopt best practices such as not recording personal identifiable information like a social security number or highly sensitive material like trade secrets for added client protection.

Tech Competence 

Using AI to transcribe client conversations triggers the duty of tech competence set out in ABA Comment 8 to Model Rule 1.1 and adopted by around 40 states. The ABA Comment doesn’t offer much guidance; it just requires lawyers to “keep abreast of changes in the law and practice,” including benefits and risks associated with relevant technology.  The benefits of using AI transcription have already been described, while the risks are readily manageable through understanding how their chosen platform works and adopting the best practices recommended below.

(JULY 2026 Update) To date, at least one regulator has weighed in on the ethics of AI recordings.  NYC Bar Formal Opinion 2025-6 concludes that lawyers must notify clients and obtain their consent before recording calls with AI-enabled tools — undisclosed recording remains deceptive under Rule 8.4 and the duty of loyalty, even if only a summary is ultimately retained, because clients speak differently when a verbatim record is being made. Under the duty of competence (Rule 1.1), attorneys must understand how these tools work (including data storage, training use, and default recording settings), independently review any AI-generated transcript or summary for accuracy — especially informal “off the cuff” advice that a written record could formalize — and consider the confidentiality, privilege, and discovery risks of preserving verbatim records under Rule 1.6. Attorneys must also train and supervise subordinates on these tools under Rules 5.1 and 5.3.

3.  Do the Terms of Service for AI-recording Platforms Provide Adequate Protection for Confidentiality?

General guidance on terms of service for AI recordings

If you plan to use AI to record client communications, you must familiarize yourself with the product’s terms of service, either by trudging through the terms yourself or using Chat GPT or another AI tool to take a first crack at a summary.  As a general rule, products designed specifically for lawyers generally comply with applicable legal and ethical requirements, but even here, there’s no substitute for reading the terms.

When it comes to choosing an AI recording tool for clients, you’ll want to avoid the following:

• Platforms or products that use the entirety of your recordings or transcriptions for training;
• Platforms or products that retain information that you’ve uploaded even after you’ve deleted it;
• Platforms that don’t take reasonable steps to protect privacy and confidentiality of your communications

Review of the Terms of Service for Fireflies.AI

With that in mind, let’s have a look at the Terms of Service and for Fireflies.AI which I’ve been using most recently.  [Beginner’s note: Be sure to review terms of service or terms of use, and not privacy policy which applies to data collected from you by the company for activities like payment or marketing, and not content you upload as part of your use of the site’s functions]. The discussion below highlights key provisions and does not go through the full terms of service which you should review in its entirety.

1.         User Content Definition

In reviewing AI terms of service, begin with the definition of ‘user content’ as that term is typically used later during a discussion of applicable protections.  Here, user content encompasses the conversations you record as well as any other materials you might upload to transcribe.

Fireflies.AI user content terms also require users to warrant compliance with applicable law.  So failing to obtain client permission in a two-party consent jurisdiction violates the site terms of service and could result in cancellation of your account.

2.         AI Provisions

The AI provisions cite risks in processing data but goes on to state that “Fireflies will take reasonable means to preserve the privacy and security of data.  For added protections, Fireflies.AI also offers a Private Cloud option, described below:

3. Other Data Protection Policies

The Terms of Service didn’t answer all my questions, so I had to dig further into Fireflies.AI’s Guide on Keeping Information Safe.  The Guide disclosed that data can be deleted completely from the system:

And also that data may be used in  a limited and anonymized fashion to improve transcription accuracy:

4.         Conclusions

Overall, the Fireflies.AIi terms and policies offer sufficient protection for most routine client communications.  That said, mileage may vary depending upon the nature of a given matter and degree of sensitivity of the information so be sure to consider your unique use case in evaluating whether to transcribe client conversations with AI and what tools to choose.

Other Recording Risks (UPDATED JULY 2026):

 When clients record.  What’s good for the goose is not always good for the gander.  Although attorneys can ethically record conversations with clients, when clients seek to record conversations, it’s a different matter entirely.  For starters, when clients record, attorneys lose control over the output.  What’s to stop a client from doctoring a recording to claim that you gave certain advice when that never happened.  Moreover, clients may select recording options that aren’t secure and could inadvertently result in waiver of attorney client privilege.

In New York, attorneys may include a provision in their engagement agreement expressly barring clients from recording conversations with attorneys.  See NY Ethics Opinion 1270.  But even if you don’t go that far, it’s important to explain to clients the risks of DIY recordings.  And if you record the conversations, you can also offer a substitute.  The Illinois Bar Association also provides tips on best practices when clients use AI notetakers.

AI Notakers It’s one thing when a client brings a notetaker.  It’s another matter when you’re conducting  multi-party settlement or negotiation and notetakers invited by various participants show up.  Here’s the problem: each party needs to consent to be recorded by each notetaker.  As meeting host, you could be on the hook if consent is lacking.  Plus, you have no control over how the recordings could be disseminated or edited later on. So be sure to block those uninvited AI scribes.

State Biometric Data Laws:  As relevant here, biometric information includes data derived from such depictions, images, descriptions, or recordings, to the extent that it would be reasonably possible to identify the person from whose information the data had been derived.  As of July 2026, several states, including Illinois, Texas, and Washington State and even local governments have adopted biometric privacy laws to protect consumers and require informed, written consent before a company may gather biometric information.  Although it’s not clear whether all statutes will apply to voice recordings, it’s best to be on the safe side and secure consent for the recording before it begins or as part of your engagement agreement.

Best Practices

1.  What Are Some Best Practices for AI-Transcription of Client Communication?

            Based on the above analysis, here’s a list of best practices that lawyers can adopt to transcribe client conversations with AI tools lawfully, ethically, and in a way that both protects, benefits, and respects client interests.

Best Practice No 1: Due Diligence in Choosing AI Transcription Tools

Apply due diligence in choosing AI transcription tools to assure confidentiality and data privacy.  Generally, tools developed specifically for lawyers will include adequate safeguards and avoid use of data for training –  but that’s still no substitute for questioning the vendor and reading the terms of service yourself.  If you plan to  use a more general recording product, review both terms of service and any additional product guidance.  If doubts about security remain, contact the vendor directly.

Best Practice No. 2:  Employ AI Transcription Tools Prudently

Use AI recording tools prudently for an added level of confidentiality.  When using a vendor platform, refrain from recording or transcribing PII or other highly sensitive information.  Also, limit your use of a third-party storage platform; instead, download transcriptions on a regular basis to your own storage, and delete the recordings on the vendor platform as soon as practicable.

Best Practice No. 3:    Disclosure to Clients

Don’t just disclose AI transcriptions to clients.  Again, explain why you record client conversations and tout the benefits and discuss the risks of DIY client recordings. In addition, you may also want to offer clients the chance to opt out.  Need a disclosure policy? Here’s Chat GPT’s first crack at it:

Best Practice No. 4: Block Unauthorized Notetakers. Monitor meetings for unauthorized AI notetakers and block them.  Advise participants in advance that note takers are not permitted. That said, because recording multiple-party meetings can be useful, agree to allow one party to record provided that they share the recording with all other parties concerned.

Best Practice No. 5: Track Applicable State Privacy Laws. When it comes to AI, lawyers must also comply with federal, state and local laws in addition to ethics rules.  Lawyers in states with biometric privacy statutes should obtain written consent for recordings and all lawyers should track privacy laws as they evolve over time insofar as AI is concerned.

Here’s a sample clause you can use as a starting point:

Use of AI Meeting Tools. To give you our full attention and an accurate record of our advice, we use secure AI tools to record, transcribe, and summarize certain client meetings and calls.  By signing this agreement, you consent to that recording, including the processing of your voice, which some state laws treat as biometric data. You may opt out at any time, for any or all conversations, with no effect on our representation. We protect these records as confidential client communications, use only vetted tools that don’t train AI on client data, avoid recording highly sensitive details, and review all transcripts and summaries for accuracy until we adopt one, it is a draft, not our legal advice. We ask you not to record with your own tools becuase consumer apps can risk waiving attorney-client privilege. We’re always happy to share our transcript with you instead so just ask.

Until now.  Last week, Crosby.AI — an AI-native law firm, not a tech vendor — released the Redline Bench, a benchmark it designed itself to claim that its attorneys outperform AI in contract negotiations.  When a law firm starts benchmarking itself and publishing the results as marketing, bar ethics rules on advertising, comparative claims, verification potentially kick in.  So the legal profession needs to understand what benchmarks are, how they work, and whether existing ethics rules are equipped to handle them.

 Benchmarks are structured evaluation datasets that measure how well an AI model performs on specific tasks.   In the legal context, that’s harder than it sounds. Most legal work produces unstructured text like briefs, opinions, contracts that resists automated scoring.  As a Stanford University study on benchmarking legal AI explains, legal tasks are often inherently subjective. A claim’s merits, a contract’s adequacy, a brief’s persuasiveness are all matters on which equally competent lawyers can reasonably disagree, which means there is no firm truth to measure a model’s output.

Accuracy notwithstanding, with hundreds of millions in venture capital flowing into legal AI, companies are under pressure to differentiate.  As a result, benchmarks have become a primary battleground in the legal tech space. Harvey released its open-source Legal Agent Benchmark (LAB), 1,250 tasks across 24 practice areas structured to mirror actual law firm assignments, graded on an all-or-nothing basis because, as Harvey explains, a deal memo that catches eight of ten risks isn’t 80% useful but materially incomplete. Thomson Reuters built CoCoBench to evaluate its next-generation CoCounsel Legal, representing over 15,000 hours of practitioner and engineering work, measuring not just whether an AI produces an answer but whether the full sequence of work behind that answer meets what TR calls a “fiduciary-grade standard.”

So what does benchmarking legal tech products have to do with lawyer ratings? That depends on who’s doing the benchmarking.

When legal tech companies like Harvey or Thomson Reuters benchmark their own AI products, the ethics picture is relatively straightforward. They’re technology vendors, not law firms, and while they answer to the Federal Trade Commission (which endorses truthful comparative advertising), bar ethics rules don’t apply.

But when a hybrid AI-native firm like Crosby.AI enters the picture, the lines blur.  Crosby, together with AI research company micro1, built the RedlineBench, a multi-turn simulated SaaS contract negotiation in which AI models were tested against rubrics authored by Crosby’s own attorneys. Those same attorneys set the standards and implicitly defined what “good” looks like. The actual scoring was done by a panel of three LLM judges evaluating model outputs against the attorney-authored rubrics. The benchmark showed that current AI models fall meaningfully short of redlining by Crosby attorneys. 

But here’s the real question. When Crosby claims that its benchmark study shows that Crosby attorneys outperform AI in contract negotiations, is it acting as a legal tech company doing marketing or as a law firm making comparative advertising claims subject to Rule 7.1?

 That difference matters under existing bar rules because performance assessments and comparative statements in attorney advertising are permissible only when substantiated by verifiable, objective data. New York Opinion 1005 said “unsurpassed litigation skills” violates Rule 7.1 because it “compares the skills of the lawyer with others without factual support” and went further, holding that even posting a disclaimer that “prior results do not guarantee a similar outcome” will not cure the ethical problem. Texas prohibits statements like “we are the toughest lawyers in town” or “we are the best law firm in Texas if you want a large recovery,” and requires that even factual claims like settlement records or damage awards be accompanied by prominent qualifications to avoid creating unjustified expectations. D.C. Opinion 249 expressed concern over a lawyer’s ad promising that “we can help you when others can’t,” calling it the kind of comparative claim that is “incapable of substantiation.”

Benchmark-based claims raise similar issues under existing rules. When Crosby.AI says its attorneys outperform AI in contract negotiations, that’s a comparative statement that runs afoul of Rule 7.1 analysis as much as a phrase as the “toughest lawyers in town.” The fact that it’s framed as a benchmark result rather than a superlative doesn’t automatically save it because as the Stanford study acknowledges, the assessments are inherently subjective.  In jurisdictions like North Carolina and New Jersey, self-laudatory awards or comparative accolades must not only disclose the methodology behind the claim but also reflect “objective, verifiable, and consistently applied factors.” Objectivity is lacking when a law firm gives its lawyers top ranking on a test it created and graded itself.

On the other hand, Crosby’s self-administered benchmark may not be as problematic as it seems. Crosby has been transparent about its methodology, disclosing who created the benchmark and how.  What’s more, Crosby’s target clients are sophisticated corporations making high-stakes decisions about when to use human lawyers and when to use technology. Those clients have in-house counsel and IT professionals who can evaluate the methodology themselves. And let’s be real: what state bar is going to take on a biglaw-adjacent, VC-backed law firm over violations of consumer marketing rules?

As always, the risk falls on solos and smalls who may seek to mimic Crosby’s benchmark approach to show clients where the value of a human lawyer (or a human lawyer empowered by AI) surpasses a machine alone.  In this context, benchmarks could prove helpful since less sophisticated consumer clients and small businesses actually need guidance to decide whether to hire a lawyer or use an AI tool far more than Crosby’s corporate clients do.  Benchmarks are far from perfect, but if fully transparent, they offer more meaningful insight to consumers than opaque ratings and rankings.   Regulators that reflexively apply decades-old advertising standards to human-versus-AI benchmark comparisons may wind up hurting the very clients they claim to protect in an increasingly AI-saturated market.

Together, Foonberg and Fan stand as diametrically opposing ends of the law firm launch continuum.  But Foonberg’s and Fan’s background and philosophy of starting a firm couldn’t be more different. 

Born in 1935, Foonberg was 41 years old (though his curmudgeonly advice made him seem much older) and mid-career when his 700-page tome hit the market, swiftly setting the record for the most frequently stolen book from law school libraries.  Foonberg’s book adopts a personal, conversational tone because he assumed his readers would do it all themselves — answering the phone, writing letters, counseling clients face to face and hustling for business.  What’s more Foonberg’s early, unflinching advice read like gospel etched in stone: Rent an office. Never let the sun set on an unreturned phone call. Do the work and don’t get paid or don’t do the work and don’t get paid.  Reputation precedes you. As a topic, technology was an after thought, best summarized in a throwaway line that it always changes. 

In contrast to Foonberg, Fan is a young woman at the inception of her legal career, just two years post LLM.  Whereas Foonberg’s conception of a firm places lawyers at the center like the sun, Fan designed a firm that could run autonomously without lawyers at all. As Fan shared, she used Open Claw to architect a firm around two specialized AI agents — a senior “Morgan” and junior “Cleo” connected to a workspace where they could accept client-style prompts, break them into subtasks, debate strategy, draft work product, and hand back recommendations. Fan set rules to ensure some human oversight – such as a mandatory requirement to elevate disagreements between Morgan and Cleo for live attorney review. 

Despite their different approaches, Foonberg and Fan have more in common than you’d expect.  For starters, as first movers in the start-a-law-firm /AI native law firm space, they’ve each amassed cult-like followings eager for every scrap of wisdom they share.  For decades, Foonberg crisscrossed the country, as a keynote speaker who dominated the solo/small firm circuit.  Meanwhile, Fan recently celebrated 10,000 new followers on LinkedIn and has spoken about agents at law schools, legal tech events and even biglaw. 

What’s more, both Foonberg and Fan are united in their commitment to the future of lawyers. Foonberg constantly reminded solos that we run businesses, not charities and that if we aren’t vigilant about collecting fees, we won’t be in business long.  Meanwhile, Fan’s vision behind the Legal AI Value Stack is that legal AI’s highest and best use (Fan initially says for legal tech but the point applies with equal force to law firms) is building AI-native firms that compete with pure automation, using AI to handle everything that scales while reserving human lawyers for the trust and accountability that software alone can never provide.  In short, the Fan model maintains a permanent seat at the table for lawyers.

Of course, while Foonberg and Fan may be the bookends, they’re far from the only game in town on how to start.  My friend Ryan McKeen of BestEra who’s both biz and tech savvy and kinda old timey sits somewhere in the middle. Next month, he’s hosting Revolutionary Era, a day long modern seminar on starting a firm that covers the six pillars (vision, operations, finance, marketing, technology, culture) that separate firms that survive from firms that scale.  And of course, there’s me, the accidental entrepreneur who started a law firm because I had no other choices and made law firm ownership my life’s work through MyShingle.com.

Throughout history, lawyers have always faced challenges, whether it’s the inability to find a job, the drudgery and disappointment of biglaw,  the competition with DIY forms, disasters like Katrina and 9/11 and COVID.  Today, the spectre of obsolescence by AI looms and while it’s scarier, it’s just another obstacle on the continuum. 

What’s constant is how we lawyers respond.  What’s constant is that our profession will always have a Foonberg, a Fan and the thousands of lawyers in-between who stared down the nose of impossible circumstances and decided to step up and build something anyway. Innovators. Creators. The Crazy Ones. The People who not only figure it out, but lead movements. 

As it always has, the future of legal belongs to the owners. 

Preliminary data backs the hype.  A recent study by Anand Shah (MIT) and Joshua Levy (University of Southern California) found that cases where parties are represented by counsel generated 23 percent more docket entries per case during the study period than they did before AI. The authors conclude that “this can be explained by attorneys using LLMs too,” which suggests that LLMs are having an important intensive-margin effect by lowering drafting costs.

But here’s the thing: are more filings really what we want?

The legal profession’s near-universal embrace of Jevons Paradox rests on an unexamined assumption that an increase in legal output is inherently beneficial. More motions. More briefs. More discovery requests. More redlining rounds.  More administrative filings. Yet quantity and value are not the same thing. If AI reduces the cost of producing legal work, the real question is not whether lawyers will generate more of it, but instead, whether they should.

Legal practice when the sky’s the limit. 

Historically, a client’s budget functioned as a form of discipline for nearly all lawyers except those at the hoity-est law firms (yes, the ones billing legal assistants out at $790/hr). Although law school trains lawyers to prize quantity (spotting more issues earns a higher grade), in practice, lawyers can only pursue as many issues as a client can afford. Cost has always forced prioritization.  Which discovery requests are worth pursuing? Which filings will actually advance the client’s interests?

AI weakens that discipline because it makes it possible to generate legal work at near-zero marginal cost.  So what does law practice look like when the sky’s the limit?

Short answer: it’s not pretty.  Take a look at some of the recent filings coming to light as a result of ongoing investigation of former immigration attorney Alexandra Lozano and her disciples for running high volume immigration law practices that allegedly steered clients into immigration applications for which they were not qualified. Services also apparently included immigration-related FOIA requests with records showing that the Ohio-based Lisinski firm filing over 10,000 requests between June and August 2025, while the Lozano firm filed more than 2500. (similar filing frenzies occurred during other timeframes). 

The missing resource: judgment

Whether every one of those requests was justified is ultimately beside the point. The sheer volume illustrates what happens when legal processes become scalable. If a FOIA request generates even a modest $200 fee, then 10,000 requests translates into $2 million in revenue in just two months. The economic incentive is obvious. The public benefit is less so.

And that, dear readers, is the real paradox behind the Jevons Paradox in law. The profession has seized on Jevons as proof that AI will not eliminate lawyers because cheaper legal work will create more legal work. But that response dodges the harder question: more work for whom, at whose expense, and to what end?

Courts, agencies, adversaries, clients, and the public all operate within systems of finite attention. Every motion must be reviewed, every discovery request answered or objected to. Every new legal filing consumes time somewhere, even if AI makes it effortless for the lawyer to generate. In that sense, AI does not eliminate cost; it simply redistributes it.

What’s even more frightening is that when lawyers no longer have budget as a natural constraint, judgment becomes the only remaining discipline. That should make us nervous.  Sad to say, Alexandra Lozano was hardly the only lawyer preaching the gospel of scale. The legal industry is awash in coaches, consultants, and influencers promising lawyers greater profits through higher volume, premium pricing, and relentless efficiency. Some teach lawyers to monetize every client interaction. Others encourage lawyers to package routine services into repeatable, high-margin products.

Against that backdrop, AI is arriving on the scene when our profession is increasingly shaped by a culture that celebrates output, growth, and revenue as ends in themselves.   In this environment, AI functions as a perverse superpower that does not simply make lawyers more efficient but instead amplifies scale for the sake of scale.  AI untethered crowds out judgment.

The Jevons Paradox may well prove true. AI may unleash a flood of new legal work. But if that work consists of redundant filings, marginal motions, bloated discovery, mass-generated administrative requests, and endless rounds of AI-assisted redlining, then we lawyers should be careful about declaring victory.

A profession saved by generating more paperwork has not been saved in any meaningful sense. It has merely found a way to keep itself busy.

The better measure of AI’s value is not how much more legal work it allows us to produce, but whether it helps us produce work that is more targeted, more strategic, more affordable, and more just. The future of law should not belong to the lawyers who can generate the most filings but to those who retain the judgment and the discipline to know when enough is enough.

And not coincidentally, that’s the space that my beloved solo and small law firms occupy.  For generations, we have lived with constraints that force us to exercise judgment, prioritize relentlessly, and focus on outcomes rather than activity for activity’s sake. In a world where AI makes unlimited legal production possible, the lawyers who thrive may not be those who can do the most, but the solos and smalls who do what matters when it matters most.

The AI consulting space is, after all, a perfect storm:  thousands of lawyers panicked about falling behind and dozens of lawyers and consultants hopped up by programs like this one promising earnings of $80k/month as an AI consultant.  In short,there’s never been a time in my recent memory that lawyers were so desperate to learn about a subject, and that so many unqualified self-proclaimed gurus were available to teach it.

But for lawyers and their clients, the risks of hiring an unqualified coach are substantial. For that lesson, let’s turn to a recent example not from the tech space, but from immigration law.

Over the past month,  news stories have emerged about Alexandra Lozano, the Tukwila immigration attorney who recently resigned from the Washington State Bar in lieu of discipline. Lozano is accused in the bar complaint of running a high-volume immigration practice that used scripted sales pitches, promised “100% protection,” pushed clients into green-card filings without confirming eligibility, charged unreasonable fees, and relied on a computer-driven case system with inadequate attorney oversight. A malpractice action filed against Lozano makes similar claims.

But here’s the part of the story that hasn’t drawn as much attention. Lozano didn’t just run this operation herself; she taught other lawyers how to use it.  In fact, as the complaint alleges, Lozano earned more than $1.7 million between 2019 and 2021 coaching other immigration attorneys on her methods. Now at least one of those disciples, the Lisinski Law Firm in Ohio, is facing its own federal lawsuit alleging it purchased Lozano’s operational model and ran the same assembly-line scheme on hundreds of Spanish-speaking clients.

The hidden truth is that coaching is an unregulated industry.  When you hire a coach or consultant, you are extending trust  professionally and sometimes financially to someone who is not bound by a licensing body, mandatory continuing education, or governing ethics rules.  What’s more, many coaching arrangements include non-isparagement clauses which means that dissatisfied participants can’t share honest reviews.  And as mentioned earlier, the lack of oversight has brutal consequences.  For Lozano’s students, buying into her program exposed them to malpractice exposure and potential bar discipline, while placing clients at risk of deportation or penalties.

Granted, the consequences of picking the wrong AI consultant aren’t as devastating as a deported client, but they’re real nevertheless. An advisor who doesn’t understand bar ethics rules around confidentiality and competence could set you up on a system that exposes client data to third-party training pipelines. One who doesn’t understand how to vet or constrain AI agents could leave your operating system vulnerable in ways you won’t discover until it’s too late.

The hard part is that due diligence here requires actually knowing what you’re looking for. So here’s a starting checklist drawn from an earlier post:

Does your AI consultant understand your bar’s ethics rules on AI? Not in the abstract — specifically. Can they point you to your jurisdiction’s current guidance? Do they understand the distinction between cloud-based tools and local models when it comes to confidentiality obligations?

Can they explain data retention and training policies for the tools they’re recommending? “It’s secure” is not an answer. “Here’s the data processing agreement and what it means for client files” is.

Have they worked specifically with law firms, or are they repurposing general business AI consulting for lawyers? The regulatory environment for legal practice is specific enough that general expertise doesn’t transfer automatically.

Do they have skin in the game? This is the question I raised twelve years ago about marketing consultants, and it still applies. Are they building practices of their own using these tools? Are they willing to stand behind outcomes, or will they pivot to “results depend on your effort” the moment something goes wrong?

The Lozano story is a cautionary tale not just about the lawyer at the center of it, but about every attorney who paid for her program without asking hard enough questions. Coaching that sounds innovative can just as easily be packaging fraud as packaging insight. In immigration law, the victims were vulnerable people fleeing desperate situations. In AI consulting, the victims are lawyers who get burned — and then their clients who bear the downstream costs.

Vet hard. Ask specific questions. And if someone’s pitch sounds more like a sales deck than a substantive program, trust that instinct.

That’s because when you rely on generic prompts or skills as input, the resulting output may not be protected by copyright – meaning that your competitor down the street can copy it, paste it onto their own website, and there’s very little you can do about it.

It’s issue that almost nobody in the AI-for-lawyers space is talking about.

Here’s the quick backstory.  As a general rule, only work created by humans qualify for copyright protection. So, what happens when AI is involved in creation.  The U.S. Copyright Office addressed that question in its January 2025 report on Copyright and Artificial Intelligence, and the conclusion was clear.  Prompts alone — no matter how detailed or how extensively iterated — don’t give humans enough control over the output to claim copyright. The Office treats prompts as “unprotectible ideas” or instructions.  In other words, you’re telling the AI what you want, but you’re not controlling how it expresses that idea. The AI is. 

The report didn’t specifically address skills, because skills weren’t around at that time. But the same logic would likely apply: if the skill is essentially a packaged set of instructions telling Claude “write a client newsletter in a friendly tone covering these five topics,” the output is going to face the same copyrightability problem as a detailed prompt. You bought instructions. Instructions don’t create copyright. The AI generated the expression.

That’s the risk inherent with plug-and-play skills and prompts.  But there’s also a solution if you take the time to create your own skills.  Here’s why.

The Copyright Office also explained that when a human author’s own creative expression is perceptible in the AI output, that expression is protectable. The Office registered a work called Rose Enigma where the artist had drawn an illustration by hand, fed it into an AI system, and the AI produced a modified version. Because the human’s original drawing was clearly visible in the output, the human got copyright protection for their contribution.

And therein lies the lesson.  If you build a skill that embeds your own expressive material — your template language, your firm’s voice and tone and unique turn of a phrase – and those elements carry through into the output, you have a solid basis for protecting your work. Not because you wrote a clever prompt or skill but because your own authored expression is sitting there in the generated work, perceptible and identifiable.

Most lawyers aren’t building skills this way because nobody has shown them how. They’re either writing prompts on the fly, buying packages that produce generic output, or giving up on AI for marketing content entirely.  So let me teach you how to create Claude skills the right way so that you and your team can generate content that you can protect from infringement.  Register for the AI for Lawyers Cohort launching May 1 here.