Vendor note. Verlo makes AI software for financial advisors, so we have an interest in this question. Our guide to AI software for financial advisors covers the wider market. We describe ChatGPT, Claude and Microsoft Copilot only from each company’s own published pages, checked on November 5 and 6, 2026 and linked in the sources at the end. Their terms change often, so read the current page before you rely on any of this. Screenshots show those pages on the dates credited and illustrate their wording, not a guarantee. This is general information, not legal advice. Have your compliance officer or securities counsel review how it applies to your firm. If something here is out of date, tell us at support@verlo.finance.
The short answer. Yes, in some setups, and not on a personal account. Regulation S-P does not ban general-purpose AI tools. It makes you responsible for any outside company that handles your clients’ information. In practice that means a business or enterprise plan with written commitments on training, retention and breach notice, a due-diligence file on the vendor, and a firm policy your team follows. Personal plans generally do not give you those commitments: ChatGPT’s personal plans share your content for model training unless you turn that off, and Claude’s personal plans use it if you allow it.
If you are an advisor, you have probably already tried it: pasted a meeting transcript into ChatGPT, asked Claude to summarize a trust document, or drafted a client email with an AI model. The useful question is not whether the tools are capable. It is whether you may put a client’s information into them, and what you owe the SEC if you do.
Why the plan matters more than the model
The same model can be sold under very different terms. OpenAI and Anthropic each publish one data policy for personal plans and another for business products, and the differences are what matter for client information: whether your content can be used for training, how long it is kept, and what admin controls and logs you get. Microsoft draws the same line between its consumer Copilot app and Microsoft 365 Copilot.
Product | Used to train models by default? | Retention and controls | Source |
|---|---|---|---|
ChatGPT Free, Plus and Pro | Yes. Data sharing is on by default, and you can turn off “Improve the model for everyone” in Data controls | Temporary Chats are not used for training and may be kept up to 30 days for safety | |
ChatGPT Business and Enterprise | No. OpenAI says that by default it does not use inputs and outputs from these plans to train its models | SSO and multi-factor authentication are listed for Business. SCIM, role-based access controls and data residency are Enterprise-only. Enterprise adds retention controls and Compliance API visibility into conversations and agent activity | |
OpenAI API | No, and not since March 1, 2023, unless you opt in | Abuse-monitoring logs are kept up to 30 days. Zero Data Retention and Modified Abuse Monitoring need OpenAI’s prior approval | |
Claude Free, Pro and Max | Only if you allow it in the model improvement setting, if a chat is flagged for safety review, or if you otherwise opt in. Incognito chats are not used | If you allow training, de-identified chats can stay in training pipelines for up to five years. Deleted chats leave back-end storage within 30 days | |
Claude Team and Enterprise | No. Anthropic says it does not use inputs or outputs from its commercial products to train models, unless you submit feedback or choose to allow it | Enterprise owners can set custom retention (30-day minimum; data is kept indefinitely until a period is set), and retention changes are logged. Anthropic’s retention article covers Enterprise only | |
Anthropic API | No, under the same commercial policy | Zero data retention agreements cover eligible Anthropic APIs and Claude Code on Enterprise plans, not the personal plans | |
Microsoft 365 Copilot | No. Microsoft says prompts, responses and data accessed through Microsoft Graph are not used to train foundation models | Microsoft says Copilot is covered by its privacy, security and compliance commitments for Microsoft 365 commercial customers |
Three cautions. A no-training default is not the same as short retention, so check both. The right-hand columns are each vendor’s own statements, not an audit of what happens, so you still need due diligence. And these pages change: several of the pages we read carry no date, so record the date you checked.
What each vendor says about your data
ChatGPT
On personal plans, OpenAI’s help center says data sharing is on by default and can be switched off in Settings under Data controls. On Business and Enterprise, OpenAI says it does not use your inputs and outputs for training by default. For the API, its documentation says the same and describes two retention options: abuse-monitoring logs are kept up to 30 days, and eligible customers can apply for Zero Data Retention or Modified Abuse Monitoring, which need OpenAI’s prior approval.

OpenAI also runs a trust portal. It lists a SOC 2 Type 2 report that covers ChatGPT Enterprise and Edu, ISO 27001 and ISO 42001 certifications and other attestations, and keeps its more sensitive documents behind a request.


Claude
Anthropic splits its policies the same way. For Free, Pro and Max, its privacy center says your chats and coding sessions are used for training if you choose to allow it, if a conversation is flagged for safety review, or if you otherwise opt in. If you allow it, de-identified chats can stay in training pipelines for up to five years. For Team, Enterprise and the API, Anthropic says it will not use your inputs or outputs to train its models by default, except for feedback you submit or data you choose to share.


Custom retention controls are an Enterprise feature, and Anthropic’s page for advisors recommends the Enterprise plan for RIAs because it adds audit logs. Anthropic’s Trust Center lists SOC 2 Type 2, ISO 27001 and ISO 42001 attestations for the API, Team and Enterprise, which you can see in our due-diligence guide.

Microsoft Copilot
Microsoft sells two products with similar names. Microsoft 365 Copilot is the work version. Microsoft’s documentation says prompts, responses and data accessed through Microsoft Graph are not used to train foundation models, and notes that models from Anthropic and OpenAI run inside Copilot as subprocessors. The consumer Copilot app has a different privacy page. Microsoft’s FAQ for the older app says it uses conversation data for training unless you opt out or fall into an excluded group, such as signed-in users with an organizational account, and says an updated app has been available since August 18, 2026 with its own privacy page. Confirm which product your staff actually use.

What Regulation S-P requires of you
Regulation S-P is the SEC’s rule on how financial firms protect and share customer information. It applies to SEC-registered investment advisers, and firms registered with a state rather than the SEC should check what applies to them. Larger firms (advisers with $1.5 billion or more in assets under management) had to comply with the amendments by December 3, 2025, and smaller firms by June 3, 2026. In November 2025 the SEC’s examination staff said they would examine compliance with the 2024 amendments. Two parts of the rule matter when you use an AI tool.
The safeguards rule makes you responsible for your vendors.
“Customer information” is broad. It means any record containing nonpublic personal information about a customer, in any form, that you hold or that is handled for you. A client’s name next to balances, holdings, a tax return, a meeting transcript or an email thread all qualify. “Sensitive customer information” is the part whose compromise could create a reasonably likely risk of substantial harm or inconvenience, such as a Social Security number, or an account number combined with a date of birth or a security answer.
An AI vendor that handles that information is likely to be a “service provider.” The rule defines the term as any person or entity that receives, maintains, processes or is otherwise permitted access to customer information through services it provides to you. In our reading, a ChatGPT or Claude account used for firm work that receives client information is a relationship the rule covers.
You need written oversight of service providers, through due diligence and monitoring, designed so each provider protects customer information and tells you as soon as possible, and no later than 72 hours, after it becomes aware of a breach of a system that holds it. When that notice arrives, you must start your incident response program.
The duty to tell clients stays with you. You can sign a written agreement under which the provider notifies affected individuals for you, but the obligation rests with you. Notice is due as soon as practicable and no later than 30 days after you become aware, unless a reasonable investigation shows the information has not been and is not reasonably likely to be used in a way that would cause substantial harm or inconvenience.
The privacy rule limits what you can share. Regulation S-P also restricts disclosing nonpublic personal information to outside companies unless an exception applies. One exception covers a company that performs services or functions for you, if you have given your initial privacy notice and have a contract that prohibits it from disclosing or using the information other than to carry out the purposes for which you disclosed it (17 CFR 248.13). Our reading is that a plan whose terms let the vendor train on your content may not meet that contract condition, which is one more reason to keep client information out of personal plans. Ask your counsel to confirm how the exception applies to each tool you use.
For the questions to ask and the evidence to collect, use our AI vendor due diligence checklist for RIAs.
Other SEC rules that apply
We found no SEC rule written specifically for advisers’ use of AI as of November 2026 (the SEC withdrew its 2023 predictive data analytics proposal in June 2025), and the SEC’s chair said in March 2026 that the Commission’s mandate “is technology neutral”. The Division of Examinations says it will check whether firms comply with “existing, technology-neutral obligations”, so the existing rules apply. The SEC has not named each rule for AI, so the list below is our reading and not an SEC statement, and a rule written for AI would change it. These are the ones that touch everyday use of ChatGPT or Claude.
Books and records (Rule 204-2)
Rule 204-2 requires advisers to make and keep specified records. Written communications relating to recommendations or advice, receipts and deliveries of funds or securities, and orders are covered (paragraph (a)(7)), as are advertisements and certain communications sent to ten or more people (paragraph (a)(11)). Records must be kept in an easily accessible place for at least five years from the end of the fiscal year of the last entry, the first two years in an appropriate office of the adviser (paragraph (e)(1)). Electronic records must be arranged and indexed so they can be found, with a duplicate copy stored separately (paragraph (g)).
We did not find SEC guidance on whether AI transcripts, summaries or chat histories are themselves required records. Law-firm commentary points toward caution:
Stark & Stark wrote in November 2024 that AI meeting transcripts and summaries would likely be treated as books and records, and recommended adopting them into the firm’s records.
Cooley wrote in April 2026 that an adviser using an AI notetaker is generating records under the rule whether or not it intends to. It recommended deciding by category of use which calls should never use an AI notetaker, and which may use one under clear retention rules, instead of judging each transcript.
Troutman wrote in June 2026 that AI output that is a required record should be captured to the same extent as human-generated records, and that keeping the prompts and the documents given to the AI can make sense for riskier uses.
The Marketing Rule (Rule 206(4)-1)
Rule 206(4)-1-1) bars an advertisement from including an untrue statement of a material fact, a material statement the adviser cannot substantiate on demand, or anything materially misleading, among other things. It applies to an advertisement however it was drafted, so anything AI writes for your website, newsletter or social posts needs the same review as anything else, especially performance and “results” language.
It also applies to what you say about AI. In March 2024 the SEC charged two advisers, Delphia (USA) Inc. and Global Predictions Inc., with false and misleading statements about their use of artificial intelligence, including Marketing Rule violations. They paid civil penalties of $225,000 and $175,000. If your marketing or filings say you use AI, make sure the claim is true and specific.
The compliance rule (Rule 206(4)-7)
Rule 206(4)-7-7) requires written policies and procedures reasonably designed to prevent violations of the Advisers Act, a review of their adequacy and effectiveness at least annually, and a designated chief compliance officer. An AI-use policy, a list of approved tools and a vendor-review file are how you show you applied it to AI.
Does removing the client’s name make it safe?
Many advisors redact names and account numbers before pasting into a personal account. It lowers the risk, but it is not a policy:
Regulation S-P’s definition turns on information about a customer, not just the name. A household’s balances, ages, employer, state and a few life events can identify it with the name removed.
Redaction is manual, so it depends on every person remembering every time, and mistakes are invisible.
It does not fix the other problems: the content still sits on a vendor’s servers under personal-plan terms, and the output you rely on may still be a record you must keep.
A workable rule is: no client information in personal accounts, redacted or not; client information only in approved business tools; and redaction as an extra layer, not a substitute. Ask counsel to set your firm’s standard for what counts as de-identified.
A safe-use checklist for your firm
Name the approved tools in writing. Say which plans the firm allows, such as a business or enterprise plan under a firm account, and prohibit personal accounts for client work.
Finish due diligence before anyone connects client data. Use the vendor checklist and keep the evidence: current attestation, subprocessor list, data terms, retention settings and breach-notice commitment.
Get the commitments in writing. No training on your data, defined retention, notice of a breach within 72 hours, and help with your own client notifications.
Turn on the controls. Training off wherever it is a setting, retention set to your policy, sharing and connectors limited, single sign-on and multi-factor authentication enforced through your identity provider, and an admin who reviews the logs.
Define what data is allowed. Decide which categories may go into which tool. Many firms bar Social Security numbers, full account numbers and credentials everywhere, and bar all client data from tools that have not passed review.
Keep a person in the loop. Review anything AI drafts before it reaches a client, check every number and name against the source, and never let AI produce performance claims, testimonials or advice you have not verified.
Decide what you keep. Set, by category of use, which outputs, prompts and logs are kept as records and where, and make sure you can retrieve them for an exam.
Train your team and re-review. Tell everyone the rules, log exceptions, and re-review each vendor at least annually and whenever its terms, subprocessors or models change.
When a purpose-built platform makes more sense
General-purpose assistants are good for research, drafting and analysis that does not involve client information. They get harder to govern when you want them to work on your clients’ data: you connect them to your CRM and custodian yourself, match their answers to the right household, and build your own audit trail. Platforms built for advisors do that connecting for you and are designed around these obligations, but they are likely to be service providers under the rule’s definition too, so the same due diligence applies. See how the options compare in our guide to AI software for financial advisors, linked at the top of this page, or book a demo to see Verlo.
Frequently asked questions
Can I put client names into ChatGPT?
We advise against it on a personal plan. On ChatGPT’s personal plans, data sharing for model training is on by default. A business or enterprise plan changes the terms, but under Regulation S-P you still need due diligence, written commitments and a policy before client information goes in. Check your firm’s policy first.
Can I use ChatGPT or Claude if I remove the client’s name?
Redaction lowers the risk but does not remove it. A household’s balances, ages and circumstances can identify it without a name, redaction depends on people doing it right every time, and the content still sits with the vendor under the plan’s terms. The safer rule is to keep client information out of personal accounts entirely and use an approved business tool for client work.
Does ChatGPT Enterprise or Claude Enterprise make my firm compliant with Regulation S-P?
Compliance is your firm’s responsibility, and a plan on its own does not discharge it. Enterprise plans can support your obligations with useful commitments and controls: no training by default, retention settings, logs and attestations such as SOC 2. Regulation S-P puts the oversight duty on you: written policies, due diligence, monitoring and your own incident response. The vendor’s plan is one input to that file, not a substitute for it.
Are my AI chat histories books and records?
It depends on what the chat contains and how you use it, and we found no SEC guidance directly on point. Written communications about advice and advertisements you send are covered by Rule 204-2, and law firms advise treating AI transcripts and summaries as potential records. Set a policy by category of use and keep what it requires.
Do I need my clients’ consent to use AI on their information?
Regulation S-P’s privacy rule has an exception for companies that perform services for you, which applies if you have given your privacy notice and have a contract limiting the provider’s use of the information. That exception does not address other laws, such as state privacy and recording-consent rules, so ask counsel what applies to your firm and your clients’ states.
What can I safely use ChatGPT or Claude for today?
Work that involves no client information: research, drafting general educational content, summarizing public articles, brainstorming and learning a new topic. Review anything you publish, because the Marketing Rule applies to advertisements however they were written.
Go deeper
AI vendor due diligence checklist for RIAs (with a downloadable DDQ)
Regulation S-P amendments: what RIAs must do about AI and other service providers
Updates to this guide
November 17, 2026: First published.
Sources
Checked November 5 and 6, 2026. Vendor pages reflect what each said on the date we read them.
Regulation S-P: 17 CFR 248.30, 17 CFR 248.13, FINRA compliance-date reminder, Goodwin alert on the $1.5 billion threshold, SEC 2026 examination priorities, November 17, 2025
Other SEC rules and statements: 17 CFR 275.204-2, 17 CFR 275.206(4)-1-1), 17 CFR 275.206(4)-7-7), SEC press release 2024-36, March 18, 2024, Chairman Atkins’ remarks, March 4, 2026, withdrawal of the 2023 predictive data analytics proposal, June 12, 2025, Division of Examinations, fiscal year 2027 priorities, October 23, 2026
Law-firm commentary: Stark & Stark, November 19, 2024, Cooley, April 24, 2026, Troutman Pepper Locke, June 23, 2026
OpenAI: Data controls in ChatGPT, ChatGPT Enterprise and Team data policy, ChatGPT Business, ChatGPT Enterprise, Enterprise admin FAQ, API data controls, Trust Portal
Anthropic: Consumer training article, consumer retention, commercial training article, Enterprise retention, zero data retention, consumer terms update, August 28, 2025, Trust Center, Claude for Financial Advisors
Microsoft: Data, Privacy, and Security for Microsoft Copilot, Copilot privacy FAQ (older app)