MorseSign inStart freeOpen Morse

Can auditors use ChatGPT? What PCAOB, AICPA, and IIA rules say about client data

The Morse teamPublished 11 min read

Yes, auditors can use ChatGPT, but as of November 2026 confidential client or employer information belongs only in an AI tool the firm or the audit function has approved, and for a CPA in public practice, only with what the AICPA code asks for first. The code bars disclosing confidential client information "without the specific consent of the client," the IIA's standards bar internal auditors from disclosing confidential information "to unauthorized parties," and PCAOB staff found that some large firms don't allow generative AI in audit procedures at all.

This isn't professional or legal advice. On November 15, 2026, we read the PCAOB's generative AI Spotlight, its Rule 3500T and QC 1000, the AICPA Code of Professional Conduct as updated through September 2026, the AICPA's October 2026 practice aid announcement and its ethics division's articles, the IIA's Global Internal Audit Standards and AI resources, a state board rule, and OpenAI's, Anthropic's, and Microsoft's own pages on training and retention. The AICPA's ethics division deserves credit. In a November 2, 2026 article, Kelly D. Mullins, communications manager for the AICPA Professional Ethics Division, wrote plainly that "Redaction does not, by itself, turn confidential client information into nonconfidential information" (Journal of Accountancy).

Does the PCAOB allow auditors to use ChatGPT?

It hasn't banned it, and it hasn't written a rule for it. The PCAOB's main statement on the subject is a July 2024 staff Spotlight that "represents the views of PCAOB staff" and "is not a rule, policy, or statement of the Board." Staff spoke with firms that together "audit the majority of the market capitalization for issuers," and found generative AI used mainly for "administrative and research activities."

The same Spotlight describes what those firms do about client data. Some "described safeguards in place that address what information can be uploaded to firms' GenAI-enabled tools in order to prevent confidential information (e.g., data of companies under audit) from becoming public." Some "currently do not allow GenAI to be used when performing audit or attest procedures due to data privacy concerns." And an engagement team member who uses such a tool "is still responsible for the results and documentation of the work."

The PCAOB's new quality control standard, QC 1000, takes effect on December 15, 2026. It doesn't mention AI, but it asks firms to keep technological resources at the "capacity, integrity, resiliency, availability, reliability, and security" their engagements need, and to weigh those resources' "susceptibility to cybersecurity breaches." A chatbot your firm never assessed sits outside that system.

Where does a public-company auditor's duty of confidentiality come from?

Mostly from the AICPA code and your state board, because the PCAOB's own ethics rule leaves confidentiality out. Rule 3500T makes firms follow the AICPA's Rule 102 on integrity and objectivity and Rule 101 on independence "as in existence on April 16, 2003." The AICPA's confidentiality rule, then Rule 301, isn't on the list. The version of Rule 3500T that takes effect on December 15, 2026 points its ethics part to EI 1000, Integrity and Objectivity, instead (PCAOB).

State rules fill the gap for licensees. Texas's 22 TAC 501.75 says a CPA "shall not voluntarily disclose information communicated to him by the client," and "Such information shall be deemed confidential." Its ten exceptions run to professional standards, PCAOB requirements, subpoenas, court orders, peer review, and board investigations. A software vendor isn't one of them.

The same rule, in subsection (d), covers lost records (22 TAC 501.75(d)). A CPA must "take all reasonable measures to maintain the confidentiality of the client records" and must notify the affected client in writing after "the loss of, or loss of control over, the confidentiality of those records." The rule defines loss to include "a cybersecurity breach or other incident exposing the records to a third party or parties without the client's consent." The rule doesn't mention AI. If a colleague has pasted client records into a personal account, ask your firm's counsel whether that sentence applies.

What does the AICPA code say about client data in AI tools?

The confidentiality rule reads, "A member in public practice shall not disclose any confidential client information without the specific consent of the client" (AICPA Code, 1.700.001). The code as updated through September 2026 adds a note under that rule pointing to a staff article, "Ethics Staff Insights: AI through an ethics lens," that "discusses threats to compliance with the code when members use AI to provide services." The article sits in the AICPA's members-only ethics library.

The code's route for third-party service providers, interpretation 1.700.040, asks for a confidentiality contract with "reasonable assurance" about the provider's procedures, or the client's specific consent. We cover how that applies to tax work in can accountants use ChatGPT. For audit firms, the AICPA's October 13, 2026 practice aid announcement says the guidance outlines engagement partners' duty to understand the AI tools they use and "ensure they're approved by the firm and are being used correctly by staff."

An AICPA ethics article from August 2026 warns that "Some publicly available generative AI tools employ user inputs to train future models," so "sensitive employer or client data that is entered could potentially resurface elsewhere" (Journal of Accountancy). It tells members to ask, before pasting client data into any AI system, "Is this confidential client information?" and "Could this use result in disclosure to third parties?"

Does removing the client's name make it safe to paste?

Not under the AICPA code. Interpretation 1.700.060 applies when client information may be "disclosed to others without the client being specifically identified." If the information is still confidential, disclosing it without "the client's specific consent, preferably in writing" violates the rule, and that consent "should specify the nature of the information that may be disclosed, the type of third party to whom it may be disclosed, and its intended use."

Mullins said the same thing about AI in that November 2, 2026 article: "Removing names and other identifying details does not change the confidential nature of the underlying information or eliminate your obligations" (Journal of Accountancy). The article was about selling firm data for model training, and it ties the point to the AI staff article and to 1.700.060.

Other guides draw the line elsewhere. One, checked against OpenAI's pages on November 4, 2026, treats an auditor's notes on a client's controls "with names, amounts and account numbers replaced" as allowed "with a coded name and the engagement partner's agreement" (Sagnik Bhattacharya). Under 1.700.060, the client's consent is the test, and the partner can't give it. Even the client's name can be confidential. The code's example is a practice limited to bankruptcy work, where naming a client "could suggest that the client may be experiencing financial difficulties" (1.700.090).

What do the IIA standards say about internal auditors using AI?

They don't name AI, but Principle 5, Maintain Confidentiality, reaches any prompt that carries what you learned on an engagement. The Global Internal Audit Standards say internal auditors receive "information that is confidential, proprietary, and/or personally identifiable" and must protect it "from unauthorized access or disclosure, internally and externally." Standard 5.1 makes them "follow the relevant policies, procedures, laws, and regulations when using information." Standard 5.2 says they "must not disclose confidential information to unauthorized parties unless there is a legal or professional responsibility to do so," and "must manage the risk of exposing or disclosing information inadvertently."

Standard 5.2 also reaches outside help. The chief audit executive "must ensure that the internal audit function and individuals assisting the internal audit function adhere to the same protection requirements." Its examples of confidential information include "individual salaries and records of personnel issues."

The IIA's own AI material is about putting AI to work. Its AI tools article describes questionnaires built on its AI Auditing Framework and a prompt guide, AI Prompting for Internal Auditors, published in August 2026. The confidentiality limits come from Principle 5 and your organization's own policies.

Internal auditors who are CPAs carry one more rule. Under the AICPA code's 2.400.070, a member in business "should maintain the confidentiality of his or her employer's confidential information," and disclosing it "without the proper authority or specific consent of the employer" violates the Acts Discreditable Rule.

What does each AI tool keep?

It depends far more on the plan than on the brand. We read each company's own page on November 15, 2026.

ToolTrains on your chats by defaultHow long chats are keptWhat your firm controls
ChatGPT Free, Plus, or ProYes. "Data sharing is enabled by default" until you turn off "Improve the model for everyone" (OpenAI)Saved chats stay until you delete them; a temporary chat "may be retained for up to 30 days for safety purposes"Nothing. It's your own account
ChatGPT BusinessNo, "by default" (OpenAI)Admins set it; deleted or unsaved chats removed "within 30 days," unless law or harm prevention requires longerAdmins "can view, access, export, and delete" chats; OpenAI staff and "specialized third-party contractors" may review for abuse
ChatGPT EnterpriseNo, "by default" (OpenAI)Admins set it; deleted chats removed "within 30 days, unless we are legally required to retain them"Retention and an audit log through the Compliance API
Claude Free, Pro, or MaxOnly if your model improvement setting allows it, then kept "for up to 5 years" de-identified (Anthropic)Deleted chats leave back-end storage "within 30 days"; flagged chats up to 2 yearsNothing. It's a consumer plan
Microsoft Copilot with enterprise data protectionNo. Prompts and responses "aren't used to train foundation models" (Microsoft)Your organization's retention policies applySensitivity labels, retention, and audit, with Microsoft as a data processor under its DPA

Microsoft says web queries Copilot sends to Bing fall under the Microsoft Services Agreement, with Microsoft acting "as an independent data controller" (Microsoft), so a firm-approved Copilot still sends a few words of your prompt to a separate service when it searches the web. A business plan answers the training question without answering the consent question. We cover ChatGPT's work plan in more depth in is ChatGPT Business private.

What can an auditor paste into AI?

We applied each rule's text, as read on November 15, 2026, to what auditors handle. "Personal AI" means a ChatGPT, Claude, or Gemini account your firm or audit function hasn't approved.

What you'd pasteThe rule that covers itWhat the rule asks firstA version for a personal AI
Trial balance, ledger detail, or a sample selection from an audit clientAICPA 1.700.001; your state board rule, like Texas 501.75The client's specific consent; in Texas, "all reasonable measures" and written notice after a loss of control"How do auditors usually test revenue cutoff at a retailer?"
A workpaper narrative with names and amounts swapped for codesAICPA 1.700.060Consent, preferably written, naming the information, the type of recipient, and the useA process described in general terms, with nothing from this client's controls
A client's name next to a concern, like a going-concern questionAICPA 1.700.090The name alone can be confidential when it reveals somethingA question about what a going-concern evaluation covers
Internal audit findings, salaries, or personnel recordsIIA Standards 5.1 and 5.2; AICPA 2.400.070 for CPAs in businessYour organization's policies, and no disclosure to unauthorized partiesA question about how to structure a finding, with no facts from the audit
Any of the above in a tool your firm approvedFirm policy; QC 1000 from December 15, 2026The firm's limits on what may be uploaded, which PCAOB staff found some firms setWhatever the firm's policy allows
PCAOB, AICPA, or IIA standards, and public filingsNone of the rules aboveNothingAll of it

If your firm hasn't said which tools are approved, ask before you assume a personal account is fine. On a firm laptop, IT can often see what you do in a browser anyway, as we explain in can my employer see my ChatGPT.

How Morse keeps an auditor's own AI questions private

Every row in that table is about the same exposure, a readable copy of your prompt sitting with an AI company the client, the firm, or the audit committee never approved, kept for as long as that company's terms allow. Morse is built so fewer readable copies exist. Its servers don't keep messages or log message content, and your chat history is stored in your browser, encrypted, not on Morse's servers.

Engagement data still belongs in the tools your firm approved, and whether anything from an audit may go into Morse is your firm's call under the rules above. Morse is for the questions that are yours: studying for the CPA or CIA exam, or reading a new standard before the firm's training catches up.

  • E2EE, for what you'd tell no one. Your browser checks the secure enclave's hardware attestation, then encrypts the message so only that enclave can read it, and Morse's servers relay ciphertext they can't read. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed. E2EE runs open-weight models like GLM, Kimi, and DeepSeek, with no web search or image generation, so look up a standard's text in a separate chat.
  • Private, for exam study and reading standards. Every new chat starts in Private, on open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok. Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to keep nothing, and where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't (what zero retention means).
  • Anonymous, for GPT, Claude, or Gemini 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 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 keep to the last column of the table.

In every mode, Morse doesn't train AI on your chats, sell your data, or show ads. Because we don't store chats, we usually have very little to give authorities (privacy policy). Clearing your browser, losing the device, or deleting your account deletes your chats, and Morse can't recover them, so keep anything you need somewhere else. For the same question in a neighboring field, see can bank employees use ChatGPT.

Try a chat from the Morse home page without an account, or start free with no card needed. E2EE chats come with Pro, at $19 a month.

Common questions

Does the PCAOB have a rule on auditors using generative AI?

Not as of November 15, 2026. Its main publication on the subject is a July 2024 staff Spotlight, which says it "is not a rule, policy, or statement of the Board." QC 1000, effective December 15, 2026, asks firms to secure their technological resources without naming AI.

If I turn off training in ChatGPT, can I paste client data?

The training switch changes what OpenAI does with your chats. It isn't the client's specific consent the AICPA rule asks for, and it doesn't put a personal account under your firm's retention or audit controls.

Is Microsoft Copilot at my firm different from free ChatGPT?

Yes. Work Copilot with enterprise data protection runs under Microsoft's Data Protection Addendum, with Microsoft as a data processor, and doesn't train foundation models on prompts. Web queries it sends to Bing fall under separate terms, with Microsoft as an independent controller.

Can I use ChatGPT to study for the CPA or CIA exam?

The confidentiality rules reach client and employer information, so a question about the wording of a standard, with nothing from an engagement in it, is outside them. Your firm's policy may still limit which tools you use on its devices.

Ask what you actually want to ask.

80+ models, characters, and images, with every chat encrypted and kept on your device. Free to start.

Start a private chat

Read next