ChatGPT for Financial Advisers: Use It Without Leaking Client Data

Can financial advisers use ChatGPT or Claude under Regulation S-P?

Yes. No SEC rule bans ChatGPT for financial advisers or any other generative AI tool. The amended Regulation S-P does require written policies and procedures that protect customer information, and that obligation follows the data into any AI tool an adviser uses. 17 CFR 248.30(a)(1) through (a)(3) requires administrative, technical and physical safeguards plus a written incident response program.

Both compliance dates have passed. Larger entities had to comply by December 3, 2025 and smaller entities by June 3, 2026, according to the SEC's small entity compliance guide. So the practical question for a CCO is narrow: when an adviser pastes a client's account details into a chatbot, does the firm control where that data goes?

Where does client data go when an adviser uses a personal ChatGPT or Claude account?

Into an account the firm does not own, under terms the firm never agreed to. A personal Free, Plus, Pro or Max account belongs to the adviser, and so do its ChatGPT data privacy settings. The firm has no admin console, no audit log, no retention setting and no contract with the vendor.

Training settings on personal accounts are a personal choice. Anthropic's 2025 consumer terms update asked Free, Pro and Max users to choose whether their chats can be used to train models, and said retention extends to five years for users who allow it. The same update states it does not apply to Claude for Work, the API or other commercial plans.

That gap matters under Reg S-P's service provider requirement. 17 CFR 248.30(a)(5) requires policies for overseeing service providers through due diligence and monitoring, including ensuring they notify the firm within 72 hours of becoming aware of a breach that results in unauthorized access to customer information they maintain. A firm cannot oversee a vendor relationship that exists only in an employee's personal account.

What changes on ChatGPT Business, ChatGPT Enterprise or Claude for Work?

The firm becomes the customer. Both major vendors publish different data terms for business plans than for consumer accounts.

• Training. OpenAI's business data page states it does not train models on organization data by default for ChatGPT Enterprise, ChatGPT Business, Edu and the API. Anthropic's privacy center says it does not use inputs or outputs from commercial products for training by default, unless a user submits feedback, which Team and Enterprise owners can turn off.
• Identity. OpenAI lists SAML single sign-on for ChatGPT Business and above, so access runs through the firm's identity provider and MFA policy.
• Retention and administration. OpenAI says qualifying organizations can configure how long business data is retained. ChatGPT Business has basic Member, Admin and Owner roles, and Enterprise adds SCIM provisioning and custom role-based access.
• Contract. A business plan comes with commercial terms the firm can review, file and reference in its vendor due diligence.

Vendor terms change. We checked the OpenAI and Anthropic pages cited here in October 2026, so record the date you reviewed them and recheck them at each annual review.

What will an SEC examiner ask about AI use?

Expect questions about policies, supervision and controls. The Division of Examinations' fiscal 2026 priorities say staff will review whether firms have adequate policies and procedures to monitor or supervise their use of AI technologies, and the security controls firms use to identify and mitigate new risks associated with AI. The same document lists oversight of third-party vendors as a Reg S-P exam focus. For the wider exam picture, see our guide to the SEC exam priorities for RIAs.

Broker-dealers have the same expectation from FINRA. Regulatory Notice 24-09 reminds member firms that FINRA's rules are technology neutral and continue to apply when firms use generative AI. It does not create new requirements, so supervision under Rule 3110 still applies; see how Rule 3110 applies to generative AI.

A good answer to an examiner is a short packet: the approved tool list, the AI section of the written policies, the vendor due diligence file, evidence that personal AI accounts are blocked, and DLP reports showing the controls run.

What should advisers keep out of AI tools, even approved ones?

Most firms draw the line at direct identifiers. A common policy lets advisers use an approved tool to draft a client letter, summarize a research report or restructure a planning memo, and prohibits pasting Social Security numbers, full account numbers, dates of birth, or a client's name together with holdings and balances.

Rules like this are easier to follow when the work stays the same. Advisers can describe a scenario ("a 62-year-old client with a concentrated stock position") instead of naming the client. The policy should also say whether AI prompts and outputs related to advice are kept as records. Rule 204-2(a)(7) covers certain written communications relating to recommendations or advice, and firms should decide with counsel how AI transcripts fit their recordkeeping. Our articles on Rule 204-2 recordkeeping and AI notetakers and off-channel communications cover the records side in more detail.

What should a CCO do this quarter?

Start with one approved platform and the controls that prove it is the only one in use. These seven steps fit in a quarter for most RIAs and broker-dealers.

1. Pick one approved platform and buy the business tier. Avoid a mix of personal subscriptions reimbursed on expense reports.
2. Connect it to your identity provider with single sign-on and MFA, and remove access the day someone leaves.
3. Block consumer AI sites and sign-ups on firm devices with web filtering, and allow only the firm's tenant.
4. Turn on DLP rules that detect Social Security numbers, account numbers and dates of birth in pastes and uploads to AI sites, starting in audit mode and moving to block.
5. Add the AI vendor to your Reg S-P service provider inventory with its contract, data terms and breach notification commitments.
6. Write a one-page AI section for your policies and procedures, train advisers on it, and collect attestations.
7. Pull a monthly report of blocked attempts and DLP hits, and keep it with your annual review file.

How does s90 help firms set up AI guardrails?

s90 sets up AI guardrails and DLP for RIAs, broker-dealers and dual-registered firms: approved AI platforms tied to your identity provider, blocks on personal AI accounts, paste and upload rules for client identifiers, and the evidence trail an examiner expects. Our agreements commit to 72-hour breach notification and 24-hour evidence production. If you want to see where your advisers are using AI today, we can start with a short review.