Client Consent to Use AI Tools on a File: When to Ask

You need client consent to use AI tools on a file when the tool creates a real risk that the client's information will be disclosed. ABA guidance says so expressly for self-learning generative AI. For other tools the test is reasonable safeguards and candid communication. Valid consent is specific, explained to the client and recorded.

This post sets out what the ABA and two bar bodies have written, what a disclosure has to contain, how an OCR service differs from a chatbot, and wording you can adapt. Ethics opinions are advisory and each state has its own rules, so check yours. This is a guide to handling documents and is not legal advice.

Do the ethics rules require client consent before a lawyer uses an AI tool?

Sometimes. Under the ABA's reading of the Model Rules, informed consent is required before a lawyer puts information relating to the representation into a self-learning generative AI tool. Outside that case, consent is not automatic, but the duty to communicate may still require you to tell the client.

The starting point is the confidentiality rule, which bars revealing client information without informed consent, implied authorization or an exception. The section on what the confidentiality rules require of a lawyer quotes the rule and its reasonable-efforts duty.

ABA Formal Opinion 512 applies that rule to generative AI, which it abbreviates GAI. Its conclusion on consent is one sentence: "because many of today's self-learning GAI tools are designed so that their output could lead directly or indirectly to the disclosure of information relating to the representation of a client, a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool." A footnote adds that the conclusion "is based on the risks and capabilities of GAI tools as of the publication of this opinion."

The opinion marks the edges of that conclusion.

The last four rows come from the communication rule, not the confidentiality rule. Model Rule 1.4 requires a lawyer, in paragraph (a)(2), to "reasonably consult with the client about the means by which the client's objectives are to be accomplished" and, in paragraph (a)(4), to "promptly comply with reasonable requests for information". Opinion 512 says that where the confidentiality rule does not require consent, "the lawyer must separately consider whether other Model Rules, particularly Model Rule 1.4, require disclosing the use of a GAI tool in the representation."

State bodies frame the trigger in their own words. The State Bar of California's practical guidance on generative AI, as revised in 2026, says "a lawyer must not input any confidential information of the client into a generative AI solution that may present material risks to confidentiality or security, absent informed client consent". Your state may draw the line elsewhere.

Why boilerplate in an engagement letter may not be informed consent

Boilerplate may fail because informed consent is defined by what the client was told, not by what the client signed. A general clause that authorizes "technology, including artificial intelligence" tells the client nothing about which information goes where or what could happen to it, so there is nothing informed about agreeing to it.

Opinion 512 says it directly: "To obtain informed consent when using a GAI tool, merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient."

The reason sits in the definition. Model Rule 1.0(e) says informed consent "denotes the agreement by a person to a proposed course of conduct after the lawyer has communicated adequate information and explanation about the material risks of and reasonably available alternatives to the proposed course of conduct." Three parts of that sentence defeat a stock clause:

• "A proposed course of conduct." The client agrees to something specific. "Any tools we find useful" is not a course of conduct.
• "Material risks." The risks have to be stated. A clause that mentions none has communicated none.
• "Reasonably available alternatives." The client has to know there is another way to do the work, and what it would cost in time or fees.

Compare two clauses for the same fictional matter.

We may use technology, including artificial intelligence tools, to assist in providing legal services to you. You consent to such use.

We would like to use [tool], a document summarizer, on the supplier contracts you sent us on 4 March. The tool's provider would receive the full text of those contracts. Its terms state that business customers' content is not used to train its models and is deleted within 30 days. The alternative is for an associate to read and summarize the contracts by hand, which we estimate at six additional hours. Please tell us which you prefer.

The second names the tool, the documents, where they go, the safeguard, the remaining exposure and the alternative. Whether it is enough in your jurisdiction is for you to judge, but it gives the client something to decide.

None of this makes the engagement letter the wrong place to raise the subject. Opinion 512 calls the engagement agreement "a logical place" for disclosures about how the firm uses these tools and for recording "any client instructions on the use of GAI in the representation." A general description of the firm's practice belongs in the letter. A consent that the rules require has to be built on a specific explanation, which may need to come later, when you know which documents and which tool are involved.

Name the terms of service you read, and the date, inside the specific explanation. Generative artificial intelligence and a conversion service can need different sentences. An engagement letter can describe the firm's ordinary practice in plain words. The duty of candor is why the later note states a real risk and a real alternative. Keep the client's reply in the matter file with that note.

What must a client be told about an AI tool for consent to count?

For consent to count, Opinion 512 says the client must have the lawyer's judgment on why the tool is being used, specific information about the risk, the kinds of client information involved, how others might use it against the client, and a clear account of the benefit. The definition in Rule 1.0(e) adds the alternatives.

The opinion's sentence in full: "For the consent to be informed, the client must have the lawyer's best judgment about why the GAI tool is being used, the extent of and specific information about the risk, including particulars about the kinds of client information that will be disclosed, the ways in which others might use the information against the client's interests, and a clear explanation of the GAI tool's benefits to the representation." It continues: "Part of informed consent requires the lawyer to explain the extent of the risk that later users or beneficiaries of the GAI tool will have access to information relating to the representation."

Turned into a checklist for a disclosure note or a conversation:

1. Why this tool, for this task. One or two sentences of your own judgment, not the vendor's description.
2. Which information. The categories of document and the kinds of detail in them: names, account numbers, medical history, trade terms.
3. Where it goes. Who receives it, and whether anyone beyond the vendor does.
4. The risk, specifically. Whether the tool learns from inputs, how long content is kept, who at the provider can see it, and what you could not find out.
5. How it could hurt. What a third party could do with the information if it surfaced.
6. The benefit. Time, cost, or a result that would not be practical by hand.
7. The alternative. How the work would be done without the tool, and what changes for the client.
8. The client's choices. Yes, no, or yes for some documents only, and how to change the answer later.

You cannot write items 3 and 4 without having read the tool's terms. Opinion 512 calls that reading a baseline for any lawyer using such a tool. The post on choosing an AI document tool that doesn't train gives six checks to run on those terms before you draft the disclosure.

Does the consent have to be in writing?

Opinion 512 does not say the consent must be written. Model Rule 1.0(b) defines "confirmed in writing" as a separate term, so read your state's confidentiality rule to see whether it uses it. Whatever the rule, write it down. A dated email in which the client replies to a specific explanation is evidence of what was said and what was agreed. A phone call is not, unless you follow it with a note to the client confirming the conversation.

Consent is not a cure for a bad tool

Some disclosures should not be put to the client at all. The D.C. Bar's Ethics Opinion 388 observes that clients "typically should not be asked to consent" to "wide ranging disclosures that could waive attorney/client privilege or otherwise make their most confidential and secret information available for third parties to see and use." If a tool's terms are that poor, choose another tool.

Does running OCR on a client's scans need the client's consent?

It depends on the service and on what the client has told you. Opinion 512's consent requirement is written for self-learning generative tools. An OCR service that does not learn from uploads and shows them to no other customer sits outside that sentence, but the confidentiality and communication duties still apply to it.

Opinion 512 is about generative AI, and it says in a footnote that many of the same concerns "are similar to the issues that exist with other AI tools and should be considered by lawyers using such technology." So the reasoning carries over even where the conclusion does not. D.C. Opinion 388 offers two questions that work for any tool, a converter included:

• "Will information I provide to the GAI be visible to the GAI provider or other strangers to the attorney-client relationship?"
• "Will my interactions with the GAI affect answers that later users of the GAI will get in a way that could reveal information I provided to the GAI?"

Put a conversion service through both.

On the second row, Opinion 512 itself points to guidance about ordinary outsourcing. It cites a Virginia ethics opinion acknowledging that there is "little purpose to informing a client every time a lawyer outsources legal support services that are truly tangential, clerical, or administrative in nature". Whether converting a scan to searchable text is clerical in that sense, for your client and your matter, is a judgment your own state's rules govern. A firm that describes its conversion practice in the engagement letter has less to argue about later.

Schema Law is a conversion service. Its site states, in its security commitments, that uploads are "never used to train or improve an AI model" and that nothing is "made visible to other customers." Those are the company's statements about itself. Read them, ask your questions, and decide where the service falls in the table for your client.

Health records. A client's consent does not replace a contract the law requires. If your firm received medical records from a hospital or health plan it represents, federal rules may require a signed agreement with any vendor that handles them, whatever the client says. The post on the business associate agreement with an OCR vendor explains when.

Sample engagement letter wording for AI and document conversion tools

Sample wording helps only if you rewrite it for your firm, your tools and your state. The clauses here are illustrations to adapt. They have not been reviewed against any jurisdiction's rules, and the general ones are disclosures, not the specific informed consent that a self-learning tool would require.

A general disclosure of conversion tools

To work with scanned and faxed documents, we use an online document conversion service that reads pages into searchable text and tables. We upload documents to it over an encrypted connection. Before approving the service we reviewed its published terms, which state that customer documents are not used to train AI models and are not visible to other customers, and that we can delete documents at any time. We delete uploaded documents when the conversion has been checked. If you would prefer that some or all of your documents not be uploaded to any outside service, tell us and we will handle them within the firm.

A general disclosure of generative AI tools

We use generative AI tools for some tasks, such as first drafts of summaries and document review. We use only tools that the firm has approved after reading their terms. We will not enter information about your matter into any tool that uses customer content to train its models unless we have first explained the specific tool, the information involved and the risks to you, and you have agreed in writing. A lawyer reviews everything a tool produces before it is used in your matter.

A clause recording client instructions

You have instructed us as follows about AI and online tools: [no restrictions / no generative AI tools on this matter / no upload of the following documents to any outside service: ...]. You may change these instructions at any time by writing to [name]. We will confirm any change in writing.

A specific consent request, sent when the need arises

Subject: Your approval needed before we use [tool] on [documents]

We propose to use [tool] to [task] on [documents, with dates]. In our judgment this is worthwhile because [reason]. The provider will receive [what]. Its terms say [exact sentence on training and retention]. What we could not confirm: [gaps]. If this information were disclosed, the main risk to you would be [risk]. Without the tool we would [alternative], which would [effect on time or fees]. Please reply "yes", "no", or "yes, except for [documents]". You can withdraw this approval at any time.

A few drafting points.

• Name what you checked and the date. "We reviewed its published terms in October 2026" ages honestly. "The service is secure" does not.
• Promise only what you control. You can promise to delete uploads when the work is done. You cannot promise what a vendor's systems do, so attribute those statements to the vendor's terms.
• Keep fees in view. Opinion 512 requires consultation where a tool bears on the fee. If a tool's cost is passed on, or its absence adds hours, say so in the clause.
• Leave room for the client's own rules. Institutional clients may send guidelines that override your letter. Add a line saying their written instructions prevail.

Recording a client's refusal and working without the AI tool

When a client says no, write down exactly what they refused, confirm it back to them, mark the matter so everyone who touches it knows, and do the work another way. A refusal is an instruction about the means of the representation, and it stays in force until the client changes it.

California's guidance is direct: "A lawyer must also follow any applicable client instructions or guidelines that may restrict or limit the use of AI in connection with the representation of the client." And ABA Formal Opinion 477R, on securing client communications, quotes the comment to Model Rule 1.6: "A client may require the lawyer to implement special security measures not required by this Rule".

A routine for the day the answer is no:

1. Pin down the scope. Ask what the refusal covers. All AI tools, or generative tools only? Online services only, or software on the firm's own computers as well? Every document, or one category?
2. Confirm in writing. Send a short email restating the instruction in the client's terms and asking them to correct anything you got wrong.
3. Mark the matter. Put a flag where staff will see it before they act: the matter's opening note, the folder name, the intake sheet. An instruction that lives only in one lawyer's inbox is easy for someone else to break.
4. Tell the team, including temporary staff and outside reviewers. The instruction binds the firm, not only the lawyer who received it.
5. Deal with what was already processed. If documents went through a tool before the refusal, tell the client which ones, delete them from the tool where you can, and note that you did.
6. Agree the alternative and its cost. Opinion 477R notes that a lawyer may need the client's informed consent on "the costs involved" when the client asks for nonstandard security measures. If manual handling adds time, tell the client before the bill does.
7. Diarize a check. Look at the matter again at a set point to confirm the instruction is being followed.

A worked example. Hollis Tran, general counsel of the fictional Brightwater Logistics, replies to a consent request: no generative AI on the Penrose dispute, but conversion of scanned invoices is fine so long as nothing is kept by the vendor. The firm's note reads:

Matter:        Brightwater v. Penrose (2026-0417)
Instruction:   No generative AI tools on any matter document.
               Scanned invoices may be converted by the firm's approved
               OCR service; uploads to be deleted once exports are checked.
Given by:      H. Tran, by email, 9 October 2026 (saved to correspondence)
Confirmed:     Reply sent same day by A. Okoye; no corrections received
Flag set:      Matter opening note and folder banner
Team told:     Associates, two paralegals, contract reviewer (email, same day)
Prior use:     None
Fee effect:    Manual summary of 40 contracts, est. 12 hours; client informed
Review:        First of each month until the matter closes

Working without the tool is ordinary work. Read the documents, build the chronology by hand, or run software that keeps files on the firm's own equipment if the client's instruction allows it. If the refusal makes a deadline or a budget unrealistic, that is a conversation to have with the client early, with the alternatives laid out, and it is the same conversation informed consent would have required in the first place.