Disclosure: Some links on this page may be affiliate links, which means we may earn a commission at no extra cost to you. Listings are independent and sponsored placements are always labeled. Learn more.
AI Vendor Due Diligence: Questions Advisors Should Ask Before Buying
Short answer: Before an AI tool touches client data, find out whether the vendor trains models on your data, how long it keeps data and where, which sub-processors and AI model providers see it, what independent security reports it has, how access is controlled, how fast it reports a breach and how you get your records out when you leave. Amended Regulation S-P requires SEC-registered advisers and broker-dealers to oversee service providers that handle customer information, including requiring breach notice within 72 hours, and FINRA reminds broker-dealers that outsourcing a function doesn't outsource responsibility for it. Ask in writing and keep the answers.
Updated October 21, 2026. Tool details come from each vendor's website and can change.
Why is AI vendor due diligence the firm's job?
Because the obligations stay with you when the work moves to a vendor. Under the SEC's 2024 amendments to Regulation S-P, covered institutions, including SEC-registered advisers and broker-dealers, must have an incident response program with written policies and procedures reasonably designed to require oversight of service providers. The SEC's small entity compliance guide says those providers must notify the firm as soon as possible, but no later than 72 hours, after becoming aware of a breach that resulted in unauthorized access to a customer information system they maintain. Larger entities had to comply by December 3, 2025 and smaller entities by June 3, 2026.
For broker-dealers, FINRA's Regulatory Notice 21-29 reminds member firms that outsourcing an activity or function to a third party doesn't relieve them of their supervisory obligations, and FINRA's Regulatory Notice 24-09 says its rules apply to generative AI, including AI features embedded in third-party products, the same way they apply to any other technology.
There's also a practical reason. AI vendors change quickly: new models, new sub-processors, new features switched on by default. A short, documented review before you buy, and a lighter check each year after, is how you notice those changes before they become your problem.
Scale the review to the risk. A tool that only ever sees public information, like a grammar checker used on marketing drafts, needs a lighter review than a notetaker that records client meetings or an assistant that reads tax returns. One approach is to sort tools into a few risk tiers based on the data they receive and whether they communicate with clients, and set the depth of diligence by tier.
Does the vendor use your data to train its AI models?
This is the first question for any AI tool, and the answer often differs by plan. Consumer tiers of general assistants may use conversations to improve models unless you opt out, while business and enterprise tiers often promise not to. Don't rely on marketing pages; get the answer from the contract or the data processing terms for the specific plan you'd buy.
Ask: Is our data, including prompts, uploaded files, recordings, transcripts and outputs, used to train or improve any model, yours or a third party's? Is that the default for our plan, and is it in the contract? If a third-party model provider processes our data, what are its training and retention terms? Can de-identified or aggregated data be used, and how is it de-identified?
Also ask about humans. Some vendors allow staff to review conversations for quality or abuse monitoring. Find out who can see your data, under what circumstances, and whether that access is logged.
How long is data kept, how is it deleted and where is it stored?
Retention. Ask how long each type of data is kept by default (recordings, transcripts, summaries, prompts, outputs, logs and backups), whether you can set shorter or longer periods, and what happens to backups after deletion.
Your own retention duties. Vendor retention isn't the same as your recordkeeping. Advisers keep certain records under Rule 204-2 and broker-dealers under SEA Rule 17a-4 and FINRA Rule 4511. If the vendor deletes a transcript after 30 days, make sure anything you're required to keep has already reached your archive. Our guide to record-keeping for AI chats and notetakers explains which outputs may be records.
Location. Ask which countries your data is stored and processed in, including by sub-processors and support teams, and whether you can restrict it.
Deletion on request. Ask how you delete a specific client's data if they ask, how long it takes and whether you get confirmation.
Which sub-processors and AI model providers touch your data?
Many AI products for advisors are built on top of a large language model from another company and run on a cloud provider. Each of those is a sub-processor with access to your data, and your diligence should reach them too.
Ask: Please send your current sub-processor list, including any AI model providers. What does each one do with our data? How will you notify us before adding or changing a sub-processor, and can we object? Do your contracts with sub-processors include the same confidentiality, training, retention and breach-notice terms you're offering us?
If the vendor can't name its model provider or won't share a sub-processor list, treat that as a significant gap and discuss it with your CCO before going further.
What does a SOC 2 report tell you, and what doesn't it?
A SOC 2 report is an examination performed by an independent CPA firm of a service organization's controls relevant to one or more of five trust services categories: security, availability, processing integrity, confidentiality and privacy. The framework is maintained by the AICPA.
What it tells you. Whether an auditor examined the vendor's controls, which categories were in scope and, in a Type 2 report, whether the controls operated effectively over a period of time. A Type 1 report covers the design of controls at a point in time.
What it doesn't tell you. A SOC 2 report isn't a certification, a guarantee or an endorsement, and it says nothing about anything outside its scope. Read the scope section: is the AI product you're buying actually covered, or only the company's older core product? Read the exceptions the auditor noted, the complementary user entity controls (the things you must do for the controls to work), and the report's period. A report that ended many months ago tells you less about today.
Ask: Can we see your most recent SOC 2 Type 2 report under NDA? Which products, systems and trust services categories are in scope? What exceptions were noted, and how were they fixed? When does the next report period end? If there's no SOC 2, what independent testing do you have, such as a penetration test summary?
Some vendors in our directory mention security programs on their own sites; for example, FINNY says it maintains an ongoing SOC 2 Type 2 examination program. Treat any vendor claim as a reason to ask for the report, not a substitute for reading it.
How are access, logins and permissions controlled?
Single sign-on and MFA. Ask whether the tool supports single sign-on with your identity provider and whether multi-factor authentication can be enforced for every user.
Roles. Can you limit who sees which clients' data, which features and which admin settings? Can an assistant see an advisor's recordings by default?
Offboarding. When an employee leaves, can you remove access immediately and keep their data in the firm's account rather than theirs?
Logs. Does the tool keep an audit log of who accessed, exported or deleted data, and can you get it? FINRA's 2026 Annual Regulatory Oversight Report describes storing prompt and output logs and tracking which model version was used as practices firms may use to monitor AI tools; ask whether the vendor makes that possible.
What should you ask notetaker vendors about recording and consent?
AI notetakers raise their own questions because they capture what clients say. Our directory lists advisor-focused options such as Jump, Zocks, Zeplyn and Finmate AI, and general business notetakers such as Fathom and Otter.ai; compare them in meeting notes & crm notes. Jump, for example, says it creates audit-ready records and can be configured to a firm's compliance policy, and Zocks highlights its approach to client privacy. Get the details in writing.
Ask: Does the tool record audio or video, or produce notes without keeping a recording? Can we turn recording off? How does the tool announce itself or capture consent at the start of a meeting? Can we customize that notice? Who owns the recording and transcript? Can we export recordings, transcripts and summaries to our archive, and in what format? Can we edit a summary, and is the original kept?
Recording-consent laws vary by state, and some require every party's consent, so whatever the tool offers, your firm's procedure should decide how consent is requested and documented. Our Jump vs Zocks comparison covers how two advisor notetakers handle recording differently.
Integrations, export and exit
Integrations. List the systems the tool will connect to (CRM, email, calendar, planning software, archive) and ask exactly what data flows each way, what permissions the integration requests and whether it writes into your CRM automatically or waits for approval.
Export. Ask what you can export, in what format and how often. Test an export during the trial, not after you've decided to leave.
Exit. Ask what happens at the end of the contract: how long you have to retrieve data, whether the vendor deletes everything afterward (including backups and sub-processor copies) and whether it will certify deletion in writing.
Lock-in. If summaries and workflows only exist inside the vendor's app, switching later is hard. Prefer tools that push the final record into systems you already control.
Incident notification and contract terms
The contract is where good answers become obligations. Before signing, have your compliance team or counsel check that it covers:
Breach notice to your firm as soon as possible and no later than 72 hours after the vendor becomes aware of a breach involving your customer information, matching Regulation S-P's service-provider requirement, with a named contact and what information the notice will include. The SEC's compliance guide notes a firm may agree in writing for a service provider to notify affected individuals on its behalf, but the firm keeps ultimate responsibility for that notice.
Data use limits: no training on your data, no sale, use only to provide the service.
Confidentiality and security obligations, including the vendor's duty to flow them down to sub-processors.
Audit and information rights: access to SOC reports and responses to reasonable security questionnaires each year.
Change notices for new sub-processors, material changes to AI models or features, and changes to data terms.
Termination assistance: export, transition time and certified deletion.
Records cooperation: help producing records if a regulator asks.
The AI vendor due diligence question list
Copy these into your questionnaire and ask for written answers. Adjust for your registration and how the tool will be used.
Company and product. What does the product do, exactly? Which features use AI, and which models power them? Who are your typical customers in financial services?
Data use. Is our data used to train or improve any model? Is that true for our specific plan, and is it in the contract? Can your staff view our data, and when?
Retention and deletion. Default retention for each data type? Configurable? How are backups handled? How do we delete one client's data?
Location. Where is data stored and processed, including by sub-processors and support?
Sub-processors. Current list, including AI model providers? How are we notified of changes?
Security assurance. Latest SOC 2 Type 2 report and scope? Exceptions noted? Penetration testing? Encryption in transit and at rest?
Access. SSO? Enforced MFA? Role-based permissions? Audit logs we can export?
Recording and consent (notetakers). Recording optional? How is consent captured? Who owns recordings and transcripts? Export formats?
Accuracy. How do you test output accuracy? How do users correct errors? Is the original output preserved?
Integrations. What data flows to and from our CRM, email and archive? Automatic or approved writes?
Incidents. How fast will you notify us of a breach? Who is the contact? Will you agree to 72 hours or less in writing?
Exit. Export formats and timing? Deletion certification? Transition support?
Records. Can the tool's outputs be captured by our archive? Will you help produce records for a regulator?
The AI vendor due-diligence prompt can turn this list into a tailored questionnaire for a specific vendor. File the answers with your approval decision, and make vendor review part of your firm's written AI policy, alongside the AI use policy prompt.
Sources
SEC: Regulation S-P amendments, final rule; Regulation S-P small entity compliance guide; Rule 204-2 (eCFR).
FINRA: Regulatory Notice 21-29; Regulatory Notice 24-09; 2026 Annual Regulatory Oversight Report, GenAI; Rule 4511.
AICPA: SOC suite of services. This guide is general information, not legal or compliance advice.
Get one practical AI-for-advisors email a week
Weekly: the newest AI tools for advisors, in 3 minutes.
New AI tools for advisors, what changed, and practical ways to use them. One email a week; unsubscribe anytime.
Confirms on our beehiiv page. Free, no spam, unsubscribe anytime. What you'll get
Tools mentioned
| Tool | Best for | Starting price | Free option | Link |
|---|---|---|---|---|
| Jump AI assistant for financial advisors that turns meetings, emails and documents into notes and workflows. | Advisors who want meeting notes plus follow-up workflows | From $100/advisor/mo | Free trial offered on the Meet plan. | Visit ↗ |
| Zocks AI assistant for financial advisors covering meeting prep, notes, onboarding and planning data. | Firms that want deep CRM integrations | From $67/user/mo (annual) | None listed | Visit ↗ |
| Zeplyn Agentic AI platform for wealth management that captures client intelligence and orchestrates workflows. | RIAs and enterprise wealth firms | Contact vendor | None listed | Visit ↗ |
| Finmate AI Agentic co-pilot for financial advisors that automates workflows and pre-fills forms. | Advisors who want admin work and form-filling automated | From $39/mo (annual) | Free trial offered. | Visit ↗ |
| Fathom AI notetaker that records, transcribes and summarizes Zoom, Google Meet and Teams calls. | Solo advisors trying an AI note-taker for internal calls | Free; paid from $15/user/mo | Free forever plan for individuals. | Visit ↗ |
| Otter.ai AI meeting agent with real-time transcription, summaries and action items. | Transcribing internal meetings and webinars | Free; Pro from $8.33/user/mo | Basic plan is free forever (300 monthly transcription minutes). | Visit ↗ |
| FINNY AI growth officer for advisors that finds money-in-motion prospects, ranks them with an F-Score and runs outreach. | Advisors who want AI-sourced prospects and automated, personalized outreach | $50/advisor/mo + 20 bps on sourced assets | None listed | Visit ↗ |
Checked against vendor websites on October 21, 2026. List prices as of October 21, 2026; prices change often, so confirm on each vendor's site.
FAQ
What questions should advisors ask an AI vendor?
Start with whether your data trains models, how long data is kept and where, which sub-processors see it, what independent security reports exist, how access is controlled, how fast breaches are reported and how you get your data out when you leave.
Is a SOC 2 report enough to approve an AI tool?
No. A SOC 2 report shows an auditor examined certain controls over a period, but you still need to check its scope, exceptions and date, and ask the AI-specific questions it doesn't cover, such as model training and sub-processors.
Does Regulation S-P apply to AI vendors?
If an AI vendor handles your customer information, amended Regulation S-P requires covered firms to oversee it as a service provider, including requiring breach notice to the firm within 72 hours.
What's the difference between SOC 2 Type 1 and Type 2?
A Type 1 report covers the design of controls at a point in time. A Type 2 report also tests whether those controls operated effectively over a period, which is usually more useful.
Do I need to redo due diligence every year?
Many firms do a lighter annual review: an updated SOC report, the current sub-processor list and any changes to data terms or AI features. Your compliance team sets the schedule.
Who should run AI vendor due diligence at a small RIA?
Usually the chief compliance officer, with help from whoever manages technology. Have counsel review the contract terms if you're unsure.