Private AI for legal questions: what to ask, what to leave out, and why it matters
The Morse teamPublished 9 min read
The safest private AI for legal questions is one where no company keeps a readable copy of your chat, and even then you should ask in general terms and leave the facts of your own case out. As of October 2026, a chat with a consumer AI app is not privileged the way a talk with your lawyer is, so whatever you type about your own dispute is a record that can be reached by the company that stores it, by a court, or by anyone who searches your device.
This isn't legal advice. It's about who can see your legal questions after you ask them. On October 20, 2026, we read the United States v. Heppner opinion, the American Bar Association's Formal Opinion 512, the State Bar of California's 2026 guidance, the Florida Bar's Opinion 24-1, and the terms, privacy policies, and help pages of Anthropic and Google. The bar guidance was written for lawyers, but it is the clearest public statement of what a confidential legal conversation requires, and the AI companies' own pages are fairly candid about what they keep.
Is it safe to ask ChatGPT legal questions?
For general questions, mostly yes. "What does a statute of limitations mean?" or "How does small claims court usually work?" reveals nothing about you, and the answer won't change if you leave your name out. The risk starts when the question is about your own matter: who did what, when, what you said, and what you're worried about. That's the kind of chat that ended up as evidence in Heppner, where a federal judge in New York held on February 17, 2026 that a defendant's chats with Claude were protected by neither attorney-client privilege nor work product. We cover the ruling, and how AI chats have reached court, in can AI chats be subpoenaed.
The AI companies say the same thing in their own words. OpenAI's usage policies bar using its services for "tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional" (OpenAI). Google's Terms of Service, effective July 30, 2026, say "don't rely on the services for medical, legal, financial, or other professional advice," and that such content "is not a substitute for advice from a qualified professional" (Google). In Heppner, when the government asked Claude whether it could give legal advice, it answered "I'm not a lawyer and can't provide formal legal advice or recommendations" (memorandum, page 8).
Does a chat with AI count as talking to a lawyer?
No. We cover the Heppner ruling in full, including the open question about AI tools a lawyer tells you to use, in can AI chats be subpoenaed. Judge Jed S. Rakoff found the AI chats lacked "at least two, if not all three" parts of the privilege test, and called the legal-advice part "a closer call," since the app itself disclaims giving legal advice (memorandum, pages 5 to 8). Sharing your lawyer's advice with the app raises a separate problem, waiver, covered next.
Anthropic's own reading of the case, in a help article dated August 26, 2026, is that the court's "reasoning turned on the consumer privacy terms and the absence of attorney direction." The same article points lawyers to commercial terms with no training (Anthropic). You and your lawyer control both of those things: which terms the tool runs under, and whether counsel directed its use.
Can pasting your lawyer's advice into a chatbot waive privilege?
The judge said so in a footnote. Heppner's lawyers argued that some of what he typed into Claude was information they had given him. Footnote 3 of the memorandum answers that "even if certain information that Heppner input into Claude was privileged, he waived the privilege by sharing that information with Claude and Anthropic, just as if he had shared it with any other third party" (memorandum, page 8).
The risk covers more than what you write yourself. A conversation with your lawyer is protected, but pasting their email into a chatbot to ask what it means can put that advice itself at risk. If you don't understand your lawyer's advice, ask your lawyer.
What do the AI companies' own terms say about legal questions?
They warn you off relying on the answers, and their privacy terms decide who else can see the question.
- Anthropic (Claude). The privacy policy, effective September 10, 2026, says "We may use your Inputs and Outputs to train and improve Anthropic AI models" unless you opt out, and chats flagged for safety review are used even if you do. It may share personal data with "government authorities, law enforcement, or other third parties" where it believes disclosure is "reasonably necessary to (i) comply with applicable law, regulation or legal process." Deleted conversations are removed from the back end "within 30 days" (Anthropic). The Consumer Terms, effective October 8, 2025, add that rating an answer lets Anthropic train on it even after you opt out (Anthropic). We go through the rest in is Claude private.
- Anthropic's usage policy for businesses that build on Claude, in a version that takes effect November 12, 2026, lists "advising an individual on their legal rights, obligations, or liability, or on what to do in a specific legal matter" as a high-risk area, where "a qualified person must meaningfully review" the AI's recommendation before it reaches the person (Anthropic). A legal-help app built on Claude has to keep a human in the loop. Your own chat on claude.ai has no such reviewer.
- Google (Gemini). The Gemini privacy hub, updated October 19, 2026, asks you: "Please don't enter confidential information that you wouldn't want a reviewer to see." Reviewed chats "are retained for up to three years" and "are not deleted when you delete your activity," and even with Keep Activity off, chats are kept "for 72 hours" (Google). More in is Gemini private.
Google deserves credit here. A company telling you not to type confidential information is about as clear as a privacy notice gets.
What do lawyers have to do before using AI on your case?
Get your consent, in many cases, and read the fine print first. The rules for lawyers show what a careful version of a legal question looks like, and you can ask your lawyer whether they follow them.
- The ABA. Formal Opinion 512, issued July 29, 2024, says that because "self-learning" AI tools can leak information from one matter into another, "a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool." It adds that "merely adding general, boiler-plate provisions to engagement letters" isn't enough, and that every lawyer should read "the Terms of Use, privacy policy, and related contractual terms" of the tools they use. Using AI for idea generation, with no client information typed in, needs no consent (ABA).
- California. The State Bar's 2026 Practical Guidance, which replaces its November 2023 version, 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." It warns that "a user may agree to the terms of use and privacy policy simply by using the product," and that breaches "could occur via both prompt content and uploaded materials" (State Bar of California).
- Florida. Opinion 24-1, dated January 19, 2024, recommends getting "the affected client's informed consent" before using a third-party AI program that would involve disclosing confidential information, and says concerns "may be mitigated by use of an inhouse generative AI" (Florida Bar).
Two of these rules apply to anyone asking a legal question. The question is safest when it carries no case-specific information at all, which is the ABA's "idea generation" carve-out. And uploaded documents count as much as typed prompts.
What should you leave out of a legal question?
Anything that ties the question to you or your case, and anything your lawyer told you. Here is what that looks like in practice.
| Leave out | Why | Ask instead |
|---|---|---|
| What your lawyer told you, quoted or pasted | Sharing it with an AI company can waive privilege over it (Heppner, footnote 3) | "What does 'without prejudice' usually mean in a settlement letter?" |
| Your name, the other side's name, and case numbers | Ties the chat to a real dispute, and to you | "A tenant and a landlord disagree about…" |
| Dates, places, and amounts that identify the event | Specific enough to find in a record search | Round figures and "last spring" |
| What you did, said, or are afraid you did | An admission written down is evidence, wherever it's stored | Ask how the rule works, not whether you broke it |
| Documents: contracts, letters, court papers, screenshots | California's bar warns breaches come "via both prompt content and uploaded materials" | Paste only the clause you need explained, with names removed |
| Your strategy, plans, or draft statements | Heppner's chats were defense strategy, written alone | Take strategy to your lawyer |
Before you send, ask whether the question would still make sense posted on a public forum under someone else's name. If it would, it's probably general enough. If not, it belongs with your lawyer.
How Morse keeps your legal questions private
Every risk above starts with the same thing. Your question sits on a company's servers, under a privacy policy that lets it train on the chat, keep it, and disclose it to "third parties," and that policy is what the court read in deciding the chats weren't confidential. Morse is built so your chats don't sit there. No app makes a chat with AI privileged, and Morse isn't a lawyer, but Morse changes who holds a copy.
- E2EE, for questions closer to your own situation, with names and identifying details still left out. Your browser checks the secure enclave's cryptographic proof before anything leaves your device, then encrypts the message so only that enclave can read it. Morse's servers relay ciphertext they can't read, and the reply is decrypted only in your browser. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed. Every E2EE reply has a lock you can open to see each check that passed.
- Private, for general questions about how the law works. It's the default for a new chat. Open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok, run with zero data retention: Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing.
- Anonymous, when you want GPT, Claude, or Gemini to explain a term. Our proxy strips your account, email, and IP address and sends the message under Morse's own account. The provider still sees the text without knowing it's you, and may keep it for a limited time to check for abuse under its own policy, so this is the mode for questions with no case details at all.
In every mode, Morse doesn't train AI on your chats, sell your data, or show ads, so there's no training switch to find. Your history is stored in your browser, encrypted, not on Morse's servers, and a document you attach is sent with the message and never stored by Morse. Because we don't store chats, we usually have very little to give authorities (privacy policy). That leaves your device as the place to protect: lock it, sign out when you step away, and use a Temporary chat, kept in memory only, for a question you don't want saved anywhere. Clearing your browser or deleting your account deletes your chats for good, and Morse can't recover them. Once a dispute is likely, deleting can be held against you: federal courts can act when electronically stored information "that should have been preserved in the anticipation or conduct of litigation" is lost (FRCP 37(e)), as we explain in can AI chats be subpoenaed, so ask your lawyer first.
On paid plans, picking an unfiltered model gets you straight legal answers that other apps refuse. A chatbot's answer still isn't a substitute for a lawyer who knows your facts.
Ask your next legal question from the Morse home page without an account, or start free. End-to-end encrypted chats come with Pro, at $19 a month.
Common questions
Can I ask AI a legal question about a friend's situation instead of my own?
Framing it as someone else's problem doesn't change who holds the chat. If the details are specific enough to identify the case, the record points to it just the same, so strip the details rather than relabel them.
Should I tell my lawyer I already asked a chatbot about my case?
Yes. Your lawyer can't protect what they don't know exists, and the chat may be a record they need to plan around. Ask before you delete anything, since deleting records once a dispute is expected can be held against you.
Is it fine to use AI to understand legal terms in a contract I was sent?
Asking what a clause like "indemnify and hold harmless" usually means is a general question and needs nothing private. Pasting the whole contract, with names, prices, and signatures, hands the company a copy of a document the other side may consider confidential.
Does a lawyer need my permission to put my case into an AI tool?
Often, yes. ABA Formal Opinion 512 says a client's informed consent is required before a lawyer inputs information about the representation into a self-learning AI tool, and that boilerplate in an engagement letter isn't enough. You can ask your lawyer which tools they use and on what terms.