Can social workers use ChatGPT? What your ethics code asks before a client's details go in

The Morse teamPublished October 31, 202610 min read

Social workers can use ChatGPT for work that contains no client information. As of October 2026, typing a client's identifiable details into a consumer chatbot shares them with a third party the client never agreed to, which the NASW Code of Ethics and the profession's 2017 technology standards restrict, and which HIPAA, FERPA, or 42 CFR Part 2 also restrict at many agencies. At least two states, Illinois and Maine, now also have laws that say when a licensed social worker may use AI for notes and what the client must be told in writing.

This isn't legal advice, and it isn't advice on social work practice. It covers privacy while you draft. We read the NASW Code of Ethics on NASW's site, the NASW, ASWB, CSWE, and CSWA Standards for Technology in Social Work Practice, NASW's June 2026 survey release, ASWB's September 2026 research announcement, Illinois Public Act 104-0054, Maine's 32 M.R.S. §7009, and the federal regulations for HIPAA, FERPA, and 42 CFR Part 2 in the eCFR, all on October 31, 2026. The profession deserves credit here. Its technology standards anticipated many of the questions AI now raises, years before ChatGPT existed.

Does the NASW Code of Ethics say anything about AI?

Not by name. As of October 31, 2026, the Code has no standard that mentions artificial intelligence. NASW said in June 2026 that its Code of Ethics Revision Workgroup "is recommending the development of supplemental guidance" on applying the Code to AI (NASW).

The Code's technology rules date from 2017, not from its 2021 amendments. NASW's Delegate Assembly approved the revisions on August 4, 2017, they took effect on January 1, 2018, and they "focused largely on the use of technology," including standards 1.03 (informed consent) and 1.07 (privacy and confidentiality) (NASW). NASW says the 2021 amendments, effective June 1, 2021, "are limited to" the Code's Purpose, the value of Integrity, and standard 1.05, Cultural Competence (NASW).

The 2017 wording already reaches a chatbot. Standard 1.07(c) says to "protect the confidentiality of all information obtained in the course of professional service," and when disclosure is justified, to disclose "the least amount of confidential information necessary." Standard 1.07(l) asks you to ensure "that clients' records are not available to others who are not authorized to have access," and 1.07(m) asks for "applicable safeguards (such as encryption, firewalls, and passwords)" on electronic communications, including "online chat sessions" (NASW Code of Ethics). A consumer AI company your client never agreed to is hard to call authorized.

Meanwhile, the profession is already using the tools. A survey of 1,179 social workers by the University of Texas at Austin's Moritz Center with NASW, run from October 2025 to February 2026, found "nearly two-thirds of respondents reported using AI in their current role," most often for "routine tasks like correspondence, reports and documentation" (UT Austin). Early advice didn't always mention privacy. NASW's own magazine, in June 2023, listed "Enter client information then ask, 'What do we want to accomplish in the next 30-60 days of treatment?'" among ways to use ChatGPT, with nothing on what to leave out (Social Work Advocates). By summer 2026, the same magazine quoted a clinician saying of free tools, "you cannot put in any identifiable information" (Social Work Advocates).

What do the technology standards ask before you use an AI tool?

That you can explain it to your client. NASW, ASWB, CSWE, and CSWA published their joint Standards for Technology in Social Work Practice in 2017, building on ASWB's 2015 model regulatory standards. Standard 3.01 says social workers "shall explain to clients whether and how they intend to use electronic devices or communication technologies to gather, manage, and store client information," and its interpretation lists what clients should know: "who will have access to it, how it will be stored, and how it will be retained." It names the risks, including that stored information "may be subpoenaed for use during legal proceedings, just as with paper records" (ASWB).

Three more lines from the same document fit AI drafting closely:

• Free tools. Standard 3.03 warns that "some free services for gathering, managing, and storing data may not be as secure as fee-for-service options," and suggests that "if identifying information is not needed, social workers can gather data on an anonymous basis so a particular client or research participant cannot be linked with the information."
• Sending outside the agency. Standard 3.07 says that before sharing information outside the organization, "social workers should ensure that they have informed consent from the client," and that information not needed should be "redacted."
• Your own phone. Standard 2.11 asks you to "consider the implications" of using personal devices for work, and says employers should ideally provide them. A chatbot app on your personal phone, signed in to your personal account, is that situation.

As of October 31, 2026, ASWB's technology resources page for regulators has no AI-specific material. On September 17, 2026, ASWB announced it is funding "the first national assessment of AI adoption and oversight among licensed social work practitioners," led by researchers at Fordham (ASWB).

Do social workers have to tell clients they use AI?

In Illinois and Maine, yes, in writing, when AI works from a recorded or transcribed therapy session. Everywhere else, the Code already expects the conversation. Standard 1.03(e) asks you to discuss "the social workers' policies concerning the use of technology in the provision of professional services." The two state laws spell out what that conversation has to cover.

Illinois. The Wellness and Oversight for Psychological Resources Act (Public Act 104-0054) covers "a licensed clinical social worker" and "a licensed social worker" who provide therapy or psychotherapy. It lets them use AI for "supplementary support," which includes "preparing and maintaining client records, including therapy notes," only where they keep "full responsibility for all interactions, outputs, and data use." When a session is "recorded or transcribed," the client must be told in writing that AI will be used and for what purpose, and must consent. Consent can't come from "the acceptance of a general or broad terms of use agreement" (Illinois General Assembly).

Maine. A section added to the social work licensing chapter by Public Law 2025, chapter 687, goes further on privacy. Before AI helps with notes from a recorded or transcribed session, the client must be told in writing "how session data collected by artificial intelligence will be stored, retained, used for training and deleted upon termination" of services. A social worker may not refuse services because a client says no, and must "ensure that any artificial intelligence technology is compliant with all state and federal confidentiality and privacy laws" (Maine Legislature).

Both laws are written around therapy and recorded sessions, so a case manager typing a summary may fall outside their letter. Maine's question is still the right test for any tool. If you can't say how the tool stores, keeps, and trains on what you type, you can't explain it to a client the way standard 3.01 asks.

What counts as de-identified for a social work case note?

It depends on which rule your setting answers to, and they don't agree. A school social worker, a hospital social worker, and a counselor at a federally funded addiction program can write the same sentence and be judged by three different tests. Here is each test, read in the regulation or standard on October 31, 2026.

Hospital and clinic social workers answer to HIPAA's Safe Harbor list of 18 identifiers (eCFR, 45 CFR 164.514), which we walk through against a sample note in can therapists use ChatGPT for notes. Social work notes carry three kinds of identifiers that a clinical note rarely has:

• Case numbers. A child welfare case ID or a Medicaid ID is a unique number, and HIPAA's list names both "health plan beneficiary numbers" and "any other unique identifying number."
• Court numbers. A docket number from family, juvenile, or housing court can lead anyone who searches that court's records to the case.
• Placements and shelters. A shelter's town, a foster home's street, or a group home's name is a place smaller than a state, and it can put someone in danger. Our guide on AI privacy for domestic violence survivors covers that risk from the survivor's side.

Can school social workers put student information into ChatGPT?

Not identifiable information, unless the district has brought the tool under its control. FERPA lets a school share records with an outside contractor without consent only when the contractor is "under the direct control of the agency or institution with respect to the use and maintenance of education records" (eCFR, 34 CFR 99.31). A personal ChatGPT account isn't. Anything you type there has to pass the school-community test in the table above.

What if your agency is covered by HIPAA or 42 CFR Part 2?

Then the rules are stricter than the ethics code. HIPAA requires written assurances from any outside company that handles identifiable client information for a covered agency. We cover that rule, which business AI plans come with a Business Associate Agreement, and why a consumer chatbot's terms of use aren't one, in can therapists use ChatGPT for notes.

Part 2 asks more. An outside service that handles substance use records must sign a written agreement acknowledging it is "fully bound" by Part 2 and that it "will resist in judicial proceedings any efforts to obtain access" to patient identifying information (eCFR, 42 CFR 2.11). The restrictions apply even when the person asking "has obtained a subpoena" (eCFR, 42 CFR 2.13). No consumer chatbot's terms make either promise. If you work at an agency, your employer's approved tools and its rules come first, and the workplace side of that is in can my employer see my ChatGPT.

How Morse keeps case drafting off a chatbot's servers

Every rule above comes back to who else holds a readable copy of what you typed. ChatGPT, Claude, and Gemini keep that copy on their servers, tied to your account, under consumer terms, and that copy is what standard 3.01 asks you to explain and what a subpoena can reach. Anthropic deserves credit for keeping Claude's incognito chats out of training even when "Model Improvement" is on (Anthropic), while Google's Gemini privacy hub asks you not to "enter confidential information that you wouldn't want a reviewer to see" (Google). We cover ChatGPT's training setting in does ChatGPT train on your data, and what each app collects beyond your words in what data do AI chatbots collect.

Morse is built so your chats don't sit on our servers at all, which changes the answers to standard 3.01's questions. Who has access: your chat history is stored only in your browser, encrypted, and in an E2EE chat our servers relay text they can't read. How it's stored and kept: on your device, locked when you sign out, and unlocking it takes two keys, one of which never leaves the device. How it's used: Morse doesn't train AI on your chats, sell your data, or show ads. And because we don't store chats, we usually have very little to give authorities.

• E2EE, for a case summary you've already stripped of identifiers. Your browser checks the secure enclave's proof that it's genuine before anything leaves your device, then encrypts the message so only that enclave can read it. The guarantee rests on the enclave hardware and attestation working as designed, and every E2EE reply has a lock you can open to see each check that passed. E2EE chats come with Pro.
• Private, for your own work with no client in it. A grant narrative, a training for new staff, a resource list for your county, or a letter template with [CLIENT NAME] placeholders. Private is the default for a new chat, and Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing.
• Anonymous, to ask GPT, Claude, or Gemini about a policy or a program without the provider knowing it's you. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider still sees the text and may keep it for a limited time to check for abuse, under its own policy, so keep every client detail out of this mode.

Identifiable client information still belongs in the systems your agency has approved. A Temporary chat is kept in memory only, which suits a draft you'll paste into the record and close. Clearing your browser deletes your chats for good and Morse can't recover them, so the finished note goes in the client's record, not in the chat. For how the encryption works, read end-to-end encrypted AI chat.

Where each kind of task goes:

• A note with a name, a case number, a school, or a placement in it: your agency's approved system.
• The same note with every identifier removed, by the strictest rule your setting answers to: E2EE.
• A grant, a training, or a template: Private, the default.
• A question about a benefits rule or a referral program for GPT, Claude, or Gemini: Anonymous.

Try a chat from the Morse home page without an account, or start free. End-to-end encrypted chats come with Pro, at $19 a month.