Deepfake and AI-Enabled Fraud Controls for Investment Advisers: What SEC Examiners Will Ask in 2027

1. The Short Answer: One Rule, Four Steps

Deepfake fraud controls for an investment adviser come down to one rule. No request by email, text, voicemail or video call moves money or changes account details until someone calls the client back on the number on file, a second person approves and the check is logged.

On October 23, 2026, the SEC's Division of Examinations released its FY2027 priorities, which list "synthetic or deepfake voice and video used for social engineering" among potential threats from SI-related cyberattacks.

The rule works whether or not anyone spots the fake. That matters, because confidence in spotting fakes is low. In a vendor survey by Tandem of 85 financial professionals at banks, credit unions and other institutions (September 14, 2026), "Only 8% of respondents expressed high confidence in employees' ability to identify AI-generated scams."

This guide is operational guidance; your counsel and CCO own the interpretation of the rules it cites. The controls are WorkWise Solutions' view of a sensible minimum, and every fact links to its source.

Staff can't be trained to hear every fake. They can be trained to dial the number on file.

2. What the 2027 Priorities Say About SI-Enabled Fraud

The priorities say "super intelligence (SI)" throughout, never AI, and never define the term. Separately, Executive Order 14434, signed September 29, 2026, told executive agencies to use SI in place of AI and defined SI as the technologies the statutory definition of AI already covers. Read SI as AI.

Four passages matter for fraud. On threats (p. 12), the document names "potential threats from SI-related cyberattacks that may affect their businesses or customers, such as SI-generated phishing messaging and synthetic or deepfake voice and video used for social engineering."

On policies (p. 12): "The Division will also prioritize examinations of policies and procedures relating to monitoring SI, including its use in fraud prevention and detection, back-office operations, compliance with AML and Reg BI obligations, and trading functions."

On identity theft (p. 13): "Examiners will assess whether firms are updating their Identity Theft Prevention Program to address emerging risks, such as those related to SI and stolen credentials." The Regulation S-ID focus is "red flags intended to prevent and mitigate against identity theft, particularly during customer account takeovers and fraudulent transfers."

On Regulation S-P (pp. 12 to 13), staff "will evaluate response program policies and procedures for unauthorized access to or use of customer information, including customer notification procedures and oversight of service providers."

Behind all four sits one standard: "whether firms continue to comply with existing, technology-neutral obligations" (p. 2). The priorities add no new rule. They ask whether your identity theft program, incident response and fraud procedures have caught up.

The priorities are staff views, and the areas they list are, in their own words, "not the only areas the Division will address in examinations." Treat them as the questions you know are coming.

3. How the Attacks Reach an Adviser

Most of these attacks end the same way: money or securities leave an account, or the details that control them change. The FBI's April 6, 2026 release on its 2025 report gives the scale: "the IC3 report features a section on artificial intelligence, which accounts for 22,364 complaints, costing Americans nearly $893 million."

• A client's voice asks for a wire. An FBI warning of December 3, 2024 says: "Criminals obtain access to bank accounts using AI-generated audio clips of individuals and impersonating them." FINRA's Protecting Your Investment Accounts From GenAI Fraud (updated December 1, 2025) adds that "fraudsters need only three seconds of audio of someone speaking to create a credible-sounding imitation".
• A principal's voice instructs staff. An FBI warning of December 19, 2025 described the impersonation of senior US officials, with requests that included "Wire funds to an overseas financial institution under false pretenses". Put your founder's voice in place of the official's and the call to your operations desk sounds the same.
• A vendor changes its bank details. An email from a known vendor, or a call in a familiar voice, says the account has moved. The next payment goes to the attacker.
• A video call looks right. FinCEN's alert of November 13, 2024 lists a third-party webcam plugin during a live verification check as a red flag. A face on a screen no longer proves who is there.
• A synthetic identity opens an account. FinCEN's other red flags include identity documents that don't match each other, a photo matching galleries of AI-generated faces, and a new account that quickly makes high-risk payments.

Clients are targets too. A joint investor alert from the SEC, NASAA and FINRA (January 25, 2024) warned: "AI technology might be used to impersonate a family member or friend, with the intent to convince an investor to transfer money or securities out of an investment account."

Two vendor surveys suggest how often it happens. Tandem's survey of 85 financial professionals at banks, credit unions and other institutions found "25% reported a known deepfake, voice cloning, or AI impersonation incident." Certos (Early Warning) surveyed 230 corporate CFOs, treasurers and controllers from August 7 to 21, 2026, with banks and credit unions excluded, and reported on October 8, 2026: "81% of corporate financial leaders surveyed say their company experienced attempted fraud involving AI-generated or AI-enhanced content in the past year."

Neither survey covers advisers, and the Certos sample is corporate finance leaders. Both suggest attempts are common among people who move money.

4. The Controls That Work

In WorkWise Solutions' view, the controls that hold up never depend on a person recognizing the fake. Every request to move money or change account details passes four steps.

Around the four steps, WorkWise Solutions recommends five supporting rules:

• No instruction changes by email alone. New bank details, a new address, phone number or email address: each change goes through the four steps, with the callback made to the details on file before the change.
• A hold on first-time payees. The first payment to a new account waits for a period the firm sets, long enough for a second check.
• Code words for principals and high-value clients. Agreed in person, kept out of email and asked for on any unusual request. The FBI gives families the same advice: "Create a secret word or phrase with your family members to verify their identities."
• Multifactor authentication on every system that can move money or change client details, so a password stolen by a phishing email isn't enough on its own.
• No exceptions for seniority. A request in the founder's voice goes through the same steps, and the founder agrees to that in writing.

A cloned voice can say anything. Answering a call placed to the client's own number is much harder.

5. Regulation S-ID Red Flags

Regulation S-ID covers SEC-registered advisers with covered accounts. Under 17 CFR 248.201(d)(1), a firm "must develop and implement a written Identity Theft Prevention Program (Program) that is designed to detect, prevent, and mitigate identity theft in connection with the opening of a covered account or any existing covered account."

The 2027 priorities ask whether that program has been updated for SI. FinCEN's alert supplies red flags to add. Two, verbatim:

• "A customer uses a third-party webcam plugin during a live verification check."
• "A customer declines to use multifactor authentication to verify their identity."

The others, paraphrased: a photo that looks inconsistent, identity documents that don't match each other, a photo matching galleries of AI-generated faces, a flag from deepfake or AI-text detection software, location or device data that doesn't match the identity documents, and a new account that quickly makes high-risk payments.

In WorkWise Solutions' view, an adviser's program should add flags from its own business:

• A contact-detail change followed soon after by a transfer request.
• An urgent wire to a first-time payee.
• A caller who resists a callback or asks staff to skip a step.
• A principal giving payment instructions by text or from a personal email account.

For suspicious activity reports that describe this fraud, the alert sets a key term: FIN-2024-DEEPFAKEFRAUD.

If your red flags were written before a voice could be cloned from three seconds of audio, they need a new page.

6. Regulation S-P: Incident Response and the 30-Day Notice

The SEC adopted amendments to Regulation S-P on May 16, 2024. Firms need an incident response program "reasonably designed to detect, respond to, and recover from unauthorized access to or use of customer information", and customer notice "as soon as practicable, but not later than 30 days" after becoming aware of an incident.

The compliance dates have passed at every size. Larger entities, including advisers with $1.5 billion or more in assets under management, had until December 3, 2025, and smaller entities until June 3, 2026, according to the SEC's small entity compliance guide.

A cloned voice that talks a staff member into reading out account details, or changing them, can amount to unauthorized access to or use of customer information. Whether a given incident triggers notice is a question for counsel; the program should say who makes that call, and how fast.

In WorkWise Solutions' view, write one deepfake scenario into the response program:

1. Contain. Stop the request, alert the custodian and any receiving bank, and lock the account's contact details.
2. Preserve. Keep the voicemail, the email with its headers, the call log and the approval record.
3. Assess. Decide with counsel what customer information was reached and whether notice is required.
4. Notify. If it is, notify within the 30 days, and record the date the firm became aware.
5. Fix. Change the control that let the request through, then test it again.

Vendors count too. The priorities name "oversight of service providers", and separately flag, in general terms (p. 6), "advisers using third parties to access clients' accounts, where existing controls may be insufficient to protect client assets and data."

The 30 days run from awareness, so the program should say who hears first and who decides next.

7. Training Staff and Preparing Clients

Train the procedure. With only 8% of Tandem's respondents expressing high confidence in employees' ability to identify AI-generated scams, detection can't be the plan.

• Rehearse the call. Each quarter, someone plays a client or principal demanding an urgent wire, and whoever answers runs the four steps.
• Give staff the words. One sentence covers it: I'll call you straight back on the number we have on file. Nobody should ever be embarrassed to say it, to a client or to the founder.
• Thank the slow-down. When someone delays a real client's wire for a callback, praise it in front of the team.
• Cover everyone who answers a phone, from advisers to assistants and operations staff.

Prepare clients the same way. Tell them in writing how you will and will not contact them: you will never ask them to move money by text, you will always call back on the number on file before acting, and you will never ask for a password or a one-time code.

The joint investor alert's warning about AI impersonating "a family member or friend" reaches your clients' own phones. A note asking them to call you on a known number before acting on any urgent request about their account gives them the same defense your staff have.

A fraudster's best tool is urgency. A rehearsed script takes it away.

8. Your Own AI Tools and Your Public Voice

FINRA's three seconds of audio changes what counts as sensitive. A principal's webinar, podcast or conference video holds far more than that, and so does a recorded client call.

Your own tools hold voice data too. If the firm records calls or uses an AI note-taker, know where the recordings and transcripts are kept, who can reach them and how long they stay.

Then look for copies nobody approved. Staff who paste client recordings or transcripts into personal AI accounts put them outside your controls, and the shadow AI guide covers how to find and fix that. If your advisers use Claude, the Cowork security guide covers client data, admin controls and audit.

If you use AI to detect fraud, expect questions about that tool as well. The priorities' examination of "policies and procedures relating to monitoring SI" includes "its use in fraud prevention and detection".

You can't take your founder off the internet. In WorkWise Solutions' view, the practical answer is to assume every principal's voice is public, and make sure no single voice can move money.

Assume the voice is already cloned. Then make it worthless.

9. Documenting It for the Exam

Examiners read records. In WorkWise Solutions' view, the exam file for AI-enabled fraud holds six things:

1. Policy. The four steps and the supporting rules, written into the Identity Theft Prevention Program and fraud procedures, with SI-related red flags added.
2. Procedures. Call scripts, the approval matrix, hold periods and who can grant an exception, ideally nobody.
3. Logs. Every money-movement and account-change request, with the callback, the approver and the time.
4. Training records. Who attended, when, and which scenario they rehearsed.
5. Tests. Simulated deepfake calls or phishing tests, the results and what changed.
6. Incident records. Every attempt, stopped or not, and how the Regulation S-P response program handled it, including service providers.

Date everything. A policy last touched before anyone could clone a voice says nothing about whether the firm has caught up.

The 2027 exam priorities guide covers the rest of the AI file, and the Form ADV guide covers what you tell clients about your own AI.

A procedure without a log is a promise. The log is the evidence.

10. A 30-Day Plan

Thirty days is enough to take a firm from an old policy to a tested control with records.

1. Week 1, map and write. List every channel where requests arrive, from the main line to advisers' mobiles, and write the four steps and the supporting rules on one page.
2. Week 2, update the programs. Add FinCEN's red flags and your own to the Identity Theft Prevention Program, and write the deepfake scenario into the Regulation S-P response program.
3. Week 3, train and tell. Rehearse the call with everyone who answers a phone, agree code words with principals and send clients the contact note.
4. Week 4, test and file. Run one simulated request end to end, review the log, fix what failed and put the evidence in the exam file.

The rule takes a month. The habit takes longer, which is why the rehearsal repeats every quarter.

11. Where to Start

Start with the request most likely to hurt: a wire to a new payee.

1. Pull your last ten wire requests and check how each was verified, and by whom.
2. Write the four steps on one page and have your principals sign it first.
3. Add the FinCEN red flags to your Identity Theft Prevention Program.
4. Book a rehearsal with everyone who answers a client call.
5. Start the log this week, even if it is a spreadsheet.

Then build the exam file from what those five steps produce.