Loading
Skip to content
Miracle Minds AI
  • Try “HIPAA”, “model risk”, “FinOps” or “agentic”.

Wealth and client-facing AI

Your regulator treats what an AI tells a client as if your firm said it. Client-facing AI pays off only when every output is supervised, retained and checked against the client's profile before it reaches them, so the control layer, not the language model, is the product.

Part of AI for Financial Services & Capital Markets. 9 sources, checked .

in SEC penalties against more than 100 firms from the off-channel recordkeeping initiative, December 2021 to fiscal 2024
$2bn+
in SEC penalties against more than 100 firms from the off-channel recordkeeping initiative, December 2021 to fiscal 2024
U.S. Securities and Exchange Commission, 2024(opens in a new tab)
of US consumers say they use AI when making financial decisions, against 63% who use financial professionals (2024)
5%
of US consumers say they use AI when making financial decisions, against 63% who use financial professionals (2024)
FINRA Investor Education Foundation, 2024(opens in a new tab)
of US consumers trusted a plausible but not necessarily accurate market projection attributed to AI, about the same as to a professional (2024)
34%
of US consumers trusted a plausible but not necessarily accurate market projection attributed to AI, about the same as to a professional (2024)
FINRA Investor Education Foundation, 2024(opens in a new tab)

The problem

FINRA's position is plain: firms are responsible for their communications whether a human or AI generated them, and chatbot messages may count as correspondence, retail or institutional communications depending on who receives them.1 Its rules are meant to be technology neutral and continue to apply when firms use generative AI.2 So an advisor copilot or client assistant inherits the same supervision, content standards and recordkeeping duties as a human advisor.

Recordkeeping is where firms have paid. Since December 2021 the SEC's off-channel communications initiative has charged more than 100 firms and produced more than $2 billion in penalties.3 Rule 17a-4 requires broker-dealers to keep communications relating to their business for at least three years, the first two in an easily accessible place.4 The SEC has also fined advisers for overstating their AI: Delphia and Global Predictions paid $400,000 combined in 2024.5

There is no AI-specific rule to wait for. The SEC withdrew its proposed predictive data analytics rule in June 2025 and said it does not intend to finalize it.6 Regulation Best Interest still requires a recommendation to a retail customer to be in that customer's best interest.7 FINRA's 2026 oversight report flags AI agents that act without human validation or beyond their intended scope.8

Clients are cautious about AI but not about plausible answers. In a 2024 FINRA Foundation experiment, only 5% of consumers said they use AI when making financial decisions, against 63% who use financial professionals. Yet about a third trusted a plausible but not necessarily accurate market projection, whether it was attributed to AI (34%) or to a professional (33%).9 A disclaimer will not catch a wrong answer. Controls on the output will.

What we build

  1. 01

    Advisor meeting copilot

    Briefs built from CRM notes, holdings and recent activity before each meeting, and summaries and follow-ups after it, with every draft captured in your books-and-records archive.

  2. 02

    Suitability check at draft time

    Compares a proposed recommendation with the client's profile, costs and reasonably available alternatives, and flags gaps for the advisor before anything is sent.

  3. 03

    Client assistant with a fixed scope

    Answers account and service questions from approved content only, routes anything that looks like a recommendation to licensed staff, and archives every exchange for supervision.

  4. 04

    Communications pre-review

    Screens AI and human drafts against FINRA content standards for balance and unsupported claims, so principals review fewer and cleaner items.

Where it goes wrong

The failure modes we design against from the first week, and the control for each.

  • The assistant starts giving advice

    A service bot that drifts into recommending products creates Reg BI obligations it was never built to meet.7 Classify intent before answering, route recommendations to people, test with adversarial prompts, and sample conversations in supervision.

  • AI output that never reaches the archive

    Drafts sent from a copilot or chatbot are business communications like any other, and firms have paid heavily for gaps in recordkeeping.4,3 Capture prompts, outputs and advisor edits in the same retention system as email.

  • Marketing claims the product cannot back

    The SEC has charged advisers for false and misleading statements about their use of AI.5 Review every AI claim in marketing against what the system actually does, and keep the evidence on file.

  • Agents acting beyond their authority

    FINRA flags agents that act without human validation or beyond the user's intended scope.8 Keep trading and money movement behind human approval, scope agent permissions tightly, and log every action.

Rules that apply

  • Regulation Best Interest, 17 CFR 240.15l-1

    Primary text on law.cornell.edu(opens in a new tab)

    Requires a broker-dealer recommending a securities transaction or strategy to a retail customer to act in that customer's best interest, through disclosure, care, conflict of interest and compliance obligations.

  • Requires preservation of originals and copies of communications relating to the business for at least three years, the first two in an easily accessible place.

  • FINRA Rule 2210 and Advertising Regulation FAQs

    Primary text on finra.org(opens in a new tab)

    Hold firms responsible for AI-generated communications and require them to be fair, balanced and supervised under the applicable communications category.

  • Confirms FINRA rules apply to generative AI and expects supervisory procedures to address technology governance, including model risk management.

How to measure it

  • Share of client-facing AI outputs archived with prompt, response and edits
  • Rate of out-of-scope or unsuitable answers found in supervisory sampling
  • Advisor time saved per client meeting
  • Suitability exceptions caught before sending versus after
  • Client complaints that involve AI-generated content

Questions buyers ask

Is a client-facing chatbot a communication under FINRA rules?

Yes. FINRA says chatbot messages may be correspondence, retail communications or institutional communications, depending on their nature and the number of recipients, and firms are responsible for them whether a person or AI produced them. You need written supervisory procedures for them, content that meets FINRA standards, and records kept like any other client communication.

Do we have to keep records of what our AI assistant tells clients?

For broker-dealers, yes. SEC Rule 17a-4 requires communications relating to the business to be preserved for at least three years, the first two easily accessible, and FINRA states that recordkeeping duties apply to AI-generated communications too. The SEC's off-channel cases produced more than $2 billion in penalties, so capture prompts, outputs and advisor edits in your existing archive from the first release.

Did the SEC's predictive data analytics rule ever take effect?

No. The SEC proposed it in 2023 and withdrew it in June 2025, stating it does not intend to issue a final rule. Any future rule would need a new proposal. Existing obligations still apply to AI used with clients, including Regulation Best Interest, the Marketing Rule for advisers and FINRA's supervision and communications rules.

Sources

Every figure on this page comes from the sources below. We checked each link on .

  1. 1.Advertising Regulation FAQs (B.4 Supervising Chatbot Communications; D.8 AI Created Communications)(opens in a new tab)FINRA · finra.org
  2. 2.Regulatory Notice 24-09: FINRA Reminds Members of Regulatory Obligations When Using Generative Artificial Intelligence and Large Language Models(opens in a new tab)FINRA, June 27, 2024 · finra.org
  3. 3.SEC Announces Enforcement Results for Fiscal Year 2024(opens in a new tab)U.S. Securities and Exchange Commission, November 22, 2024 · sec.gov
  4. 4.17 CFR 240.17a-4 Records to be preserved by certain exchange members, brokers and dealers(opens in a new tab)Legal Information Institute, Cornell Law School (Code of Federal Regulations text) · law.cornell.edu
  5. 5.SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence(opens in a new tab)U.S. Securities and Exchange Commission, March 18, 2024 · sec.gov
  6. 6.Withdrawal of Proposed Regulatory Actions (Release No. 33-11377)(opens in a new tab)U.S. Securities and Exchange Commission, June 17, 2025 · sec.gov
  7. 7.17 CFR 240.15l-1 Regulation best interest(opens in a new tab)Legal Information Institute, Cornell Law School (Code of Federal Regulations text) · law.cornell.edu
  8. 8.2026 FINRA Regulatory Oversight Report: GenAI: Continuing and Emerging Trends(opens in a new tab)FINRA, December 9, 2025 · finra.org
  9. 9.The machines are coming (with personal finance information). Do we trust them?(opens in a new tab)FINRA Investor Education Foundation, May 2024 · finrafoundation.org