Skip to content

Privilege, and software that can read your mail: a checklist for attorneys

On this page

Yes, it’s ethical for attorneys to use an AI email assistant, as long as you can still meet the rules you already work under. The ABA said so in Formal Opinion 512 on July 29, 2024. The opinion doesn’t create a new rule for AI. It walks through competence, confidentiality, communication, supervision and fees, and asks you to meet each one with the tool in the room. A private email assistant for attorneys is one you can vet against all five, from the vendor’s own documents, before a client’s mail goes in.

This page is the checklist for doing that as an individual lawyer.

  • Know the tool well enough to explain it. Opinion 512 asks for “a reasonable understanding” of what it can and can’t do. You don’t need to be an expert.
  • Privilege now has a case on point. In February 2026 a federal court held that a client’s chats with a public AI tool weren’t privileged. The court read the tool’s privacy policy to decide it.
  • Keep sending in your hands. For client replies, the setting that matters most is the one that makes a draft wait for you.
  • You own every word that goes out. The opinion says the lawyer “is fully responsible for the work,” whatever level of review you choose.

Is it ethical for attorneys to use an AI email assistant?

It is, and the rules that decide it are ones you know. Opinion 512 opens by listing them. Lawyers using these tools “must fully consider their applicable ethical obligations, including their duties to provide competent legal representation, to protect client information, to communicate with clients, to supervise their employees and agents,” and to charge reasonable fees.

Here’s what each rule asks once an assistant reads your mail.

RuleWhat it asks of youWhere it shows up in your inbox
1.1, competenceUnderstand the tool, and check its workReading a summary before you advise from it
1.6, confidentialityReasonable efforts to keep client information from people who shouldn’t have itWhat the vendor keeps, trains on and discloses
5.3, supervisionReasonable efforts so outside services fit your obligationsVetting the vendor, and training your staff
1.4, communicationTell clients what they need to make decisionsWhether and how you mention the assistant
1.5, feesBill reasonablyBilling for time the assistant saved

Your state’s version of each rule is the one that binds you. Many state bars have issued their own AI guidance as well, so check yours before you rely on the ABA’s.

Technology competence

Comment [8] to Rule 1.1 carries the technology clause:

To maintain the requisite knowledge and skill, a lawyer should keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology

Opinion 512 sets the bar for AI at a sensible height. Lawyers “need not become GAI experts.” They must have “a reasonable understanding of the capabilities and limitations” of the tools they use. For an email assistant, that comes down to three things you can say in plain words:

  • what it reads, and what it can’t reach
  • what it keeps, and for how long
  • what it does on its own, and what waits for you

If you can explain those to a client in a minute, you’ve met the bar. If you can’t, the vendor’s terms and settings page are the place to start.

Rule 1.6, and what the vendor can do with what it reads

Rule 1.6(c) is the duty underneath the rest:

A lawyer shall make reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of a client.

An assistant has to read mail to sort it, summarize it or draft from it. So the question under 1.6 is what happens after the reading. Opinion 512 turns that into a consent question for “self-learning” tools. The full treatment, with the opinion’s text on consent and on vendor terms, is in what a private email assistant for law firms reads and keeps.

Rule 5.3 adds the vendor itself. Comment [3] names “using an Internet-based service to store client information” as a service you must make reasonable efforts to supervise. Opinion 512 applies the same outsourcing checks to AI tools: “These concepts also apply to GAI providers and tools.”

What happens to privilege when software reads the mail

Confidentiality and privilege are different duties. Comment [3] to Rule 1.6 draws the line. Privilege is a rule of evidence about what can be compelled from you. Confidentiality covers “all information relating to the representation, whatever its source.” An assistant touches both. Rule 1.6 asks whether you took reasonable care. Privilege asks a narrower question later, in front of a judge: was this communication kept confidential?

Until this year that second question had no AI case. Now it has two.

United States v. Heppner. On February 17, 2026, Judge Jed S. Rakoff of the Southern District of New York explained a ruling from the bench a week earlier. A criminal defendant had used Claude, on his own, to work through his legal position. He then shared the results with his lawyers. The court held the exchanges weren’t privileged and weren’t work product (memorandum, ECF No. 27).

The confidentiality finding is the part that matters for you. The court read the privacy policy the defendant had agreed to:

This is not merely because Heppner communicated with a third-party AI platform but also because the written privacy policy to which users of Claude consent provides that Anthropic collects data on both users’ “inputs” and Claude’s “outputs,” that it uses such data to “train” Claude, and that Anthropic reserves the right to disclose such data to a host of “third parties,”

The court also left a door open. The defendant acted without his lawyers’ direction, and that mattered:

Had counsel directed Heppner to use Claude, Claude might arguably be said to have functioned in a manner akin to a highly trained professional who may act as a lawyer’s agent within the protection of the attorney-client privilege.

That’s a reference to United States v. Kovel, the doctrine that covers accountants and other agents a lawyer brings in to help give advice.

Warner v. Gilbarco. The same week, on February 10, 2026, a magistrate judge in the Eastern District of Michigan went the other way on work product. A plaintiff representing herself had used ChatGPT to draft filings. The court held those materials were protected, calling AI programs “tools, not persons” (docket, No. 2:24-cv-12333). The court reasoned that waiving work product takes disclosure to an adversary, and a tool isn’t one.

Neither case is about a lawyer’s own email assistant. Both concern a litigant’s use of a public chatbot. Read together, they still point the same way for you:

  • The terms are evidence. A court asking whether mail stayed confidential may read what the vendor’s policy lets it do with that mail. Training rights and broad disclosure rights are what sank the claim in Heppner.
  • Direction counts. A tool you chose, configured and supervise as part of the representation sits closer to Kovel than a client’s private experiment.
  • Your client’s own AI use is a new risk. A client who pastes your advice into a public chatbot may have handed the other side an argument. Say so in your engagement conversation.

That last point belongs with the client-side risks you already warn about. A client writing from an employer’s account, forwarding your advice, or copying a third party can lose privilege the same way. Those are covered in is emailing a lawyer confidential?

Vetting a private email assistant for attorneys

Work through this once per tool, and keep the answers on file. A dated copy of each document is your record of reasonable efforts if anyone asks later.

  1. Training. Find the sentence about training models on your content. Note whether it covers the vendor’s own models, its AI providers’ models, or both. If the tool trains a shared model on your mail, Opinion 512 calls for client consent first.
  2. Disclosure. Read what the vendor may disclose, and to whom. Look for a promise to notify you of a subpoena or a breach. Opinion 512 lists that notice among its vendor checks.
  3. Retention. Find out how long mail, summaries and search indexes are kept. Ask what happens to them when you leave.
  4. Human access. Find out who at the vendor can read your mail, and when.
  5. AI providers. Get the list of companies that process excerpts, and the terms they process them under.
  6. Regulated data. Check whether the standard terms exclude health information or other regulated data. If you hold protected health information for a client, you may need a business associate agreement.
  7. Your clients’ rules. Read the outside counsel guidelines of any corporate client whose matters run through the mailbox. Some now speak to AI.
  8. Your state. Check your bar for an AI opinion, and note anything it adds to the ABA’s list.

The longer version of the vendor questions, written for any industry, is in what to check before connecting your inbox.

Settings to leave off, and which mailbox to connect

The vetting tells you what a tool can do. The settings decide what it does in your practice. These are the ones worth a deliberate choice on day one.

  • Training and “help improve the product.” If the tool offers a toggle to share your content for model improvement, leave it off for any mailbox with client mail.
  • Sending on its own. Keep client replies at the setting where a draft waits for your review. An email that reached opposing counsel can’t be called back from their server.
  • Automatic forwarding and rules that act on mail. An assistant that can create forwarding rules or move mail out of the mailbox should do it only with your approval.
  • Other apps with access. Review what else holds a grant to the mailbox. An old plug-in with read access is a disclosure path you’ve stopped thinking about.
  • Personal AI accounts. A free chatbot account on a lawyer’s phone, used to “just tidy up” a client email, is the Heppner fact pattern. Your firm policy should name the tools staff may use.

Then choose the mailbox. Connect your own mailbox first, and leave shared ones out until you’ve decided who should see them. An assistant connected to a shared intake address or a partner’s delegated inbox sees every client in it. Anyone with a seat on that assistant sees the summaries too. If a lateral hire is screened from a matter, they shouldn’t have access to any mailbox that carries it.

Checking what the assistant writes

Opinion 512 is direct about output:

Because GAI tools are subject to mistakes, lawyers’ uncritical reliance on content created by a GAI tool can result in inaccurate legal advice to clients or misleading representations to courts and third parties.

It also says how much checking is enough. That depends “on the GAI tool and the specific task that it performs.” Its example is a tool summarizing contracts. A lawyer who tested it against a subset reviewed by hand need not re-read every contract. The same logic works for email. Check a week of summaries against the originals. Then you know where the tool is reliable and where it isn’t.

What to look at, in practice:

  • Summaries before advice. A summary tells you what a thread is about. Read the message itself before you advise on it, quote it or calendar from it.
  • Dates and deadlines. Check any date the assistant pulled out against the message it came from. A hearing date is the wrong place to trust a paraphrase.
  • The recipient line. Confirm every address on a draft, especially on a new thread. Autocomplete picks the wrong Sarah as easily for software as for people.
  • Facts from another matter. An assistant drafting from your own correspondence can carry a detail from one client’s thread into another’s reply. Read for anything you didn’t expect to see.
  • Tone with an adversary. A draft to opposing counsel should sound like you on a careful day. Change what doesn’t.

The opinion also draws a line no setting can move. Lawyers “may not leave it to GAI tools alone to offer legal advice to clients, negotiate clients’ claims, or perform other functions that require a lawyer’s personal judgment.” A reply that gives advice or concedes a point is yours to write, or at least yours to read every word of.

Footnote 52 adds a small habit worth keeping. It suggests marking AI-produced material “when stored in any client or firm file so future users understand potential fallibility of the work.” A note in the file is enough.

Billing for time the assistant saved

Opinion 512 applies the hourly rules plainly. If you bill by the hour, you bill the time you actually spent. That includes the time you spent reviewing the draft. If a reply took four minutes to check instead of twenty to write, the client pays for four.

The opinion adds one more line. A lawyer “may not charge a client to learn about how to use a GAI tool or service that the lawyer will regularly use for clients.” Learning your email assistant is overhead.

What to tell clients

Two separate duties apply here. Rule 1.6 may require informed consent, depending on what the tool does with client information. Rule 1.4 may require you to explain your use even where consent isn’t needed. Either way, Opinion 512 says boilerplate isn’t enough:

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.

When consent is owed and what a plain explanation sounds like is the subject of what to tell clients about an AI email assistant. Two lines to add to the conversation you already have about email:

  • how your firm uses an assistant on correspondence, and that a lawyer reviews what goes out
  • a request that the client keep your advice out of public AI tools, with Heppner as the reason

How Point works in a lawyer’s mailbox

Point is an AI email client that sits on the Gmail or Microsoft 365 account you already have. Point reads the mail, sorts it, briefs you each morning and drafts replies. Here’s how that maps onto the checklist above.

  • Training. Point’s terms, version 1.0 effective July 7, 2026, say at section 6.8 that Point won’t use your content to train generalized third-party or foundation models unless you expressly authorize it. Retention, human access and the AI providers Point uses are set out in what a private email assistant for law firms reads and keeps.
  • Drafts wait for you. How far Point acts on its own is set for each kind of action, and every kind starts on review. A client reply is written and waiting when you open the thread. You read it and press send. You can raise a setting later, and at the top of the dial Point acts without asking. For client replies, review is the setting to keep.
  • A draft is marked. A gray auto-drafted mark shows Point wrote it. Change a line and the mark comes off, because the words are now yours.
  • A record of what happened. Every action Point took is in the activity log, with a time on each line.
  • Strangers wait at the door. Mail from a sender with no history is held before Point reads it into anything. You can open it in a sealed view where nothing runs. When an email is written for the AI and not for you explains why that matters.

Three lines in the terms matter for a law firm. Section 5.1 leaves attorney-client privilege and client confidentiality with you. The same section excludes HIPAA health information, and data under professional-secrecy rules Point hasn’t expressly accepted, unless a separate written supplement covers it. Ask Point in writing whether your practice needs one before you connect a mailbox.

Point also keeps separate businesses you run apart from each other. That’s a line between organizations, and it isn’t an ethical wall between clients in one mailbox. The difference is set out in keeping one business isolated from another.

The rest of what Point does is on the benefits page.

Common questions

Is it ethical for attorneys to use an AI email assistant?

Yes, if you meet the rules you already work under. ABA Formal Opinion 512 (July 29, 2024) applies the existing duties of competence, confidentiality, communication, supervision and reasonable fees to AI tools. In practice that means understanding the tool, vetting the vendor’s terms, keeping client replies on review, and checking the output before you rely on it.

Does using an AI email assistant waive attorney-client privilege?

No court has ruled on a lawyer’s own email assistant. In United States v. Heppner (S.D.N.Y., February 2026), a client’s own chats with a public AI tool weren’t privileged. The court relied partly on a privacy policy that allowed training and disclosure to third parties. It suggested a tool used at counsel’s direction might be treated like a lawyer’s agent. Choose a tool whose terms bar training and limit disclosure, and use it as part of the representation.

What settings should a lawyer turn off in an AI email assistant?

Leave off any option to share content for model training. Keep client replies on review so nothing goes out unread. Require approval for forwarding rules and for moving mail out of the mailbox. Remove old apps that still hold access, and connect your own mailbox before any shared one.

How closely do I have to check what the assistant writes?

Opinion 512 says it depends on the tool and the task. Test the tool first, for example by checking a week of summaries against the originals. Always read the underlying message before you advise from a summary. Read every word of a reply that gives advice. The opinion says you’re “fully responsible for the work” whatever level of review you choose.

Can I bill a client for time the assistant saved?

No. If you bill hourly, Opinion 512 says you bill your actual time, including the time spent reviewing the draft. You also can’t bill a client for learning a tool you’ll use regularly across clients.

Sometimes. Opinion 512 requires informed consent before client information goes into a “self-learning” tool whose output could disclose it. Where consent isn’t required, Rule 1.4 may still call for an explanation, and you must disclose if a client asks. A boilerplate clause in the engagement letter isn’t enough for consent. The details are in what to tell clients.

Ready for a calmer inbox?

Join the private beta

We're onboarding a few teams at a time. Leave your email, confirm it once, and we'll send an invitation the moment a place opens.

By joining you agree to our privacy policy.

Private beta

What you're joining

It runs on the mail you have

Point sits on top of Gmail or Outlook. Your address, your history and your contacts stay exactly as they are, so there is nothing to migrate.

You set how much Point does

Out of the box everything waits for your review, replies included. You hand over only what you trust, one kind of work at a time.

Join the private beta

We're onboarding a few teams at a time. Leave your email, confirm it once, and we'll send an invitation the moment a place opens.

By joining you agree to our privacy policy.