AI and customer data privacy: what small businesses must consider
What to think about before customer data touches an AI tool: what the FTC has said about data promises, how the NIST AI Risk Management Framework helps, a vendor question list, and an 11-step checklist. General information, not legal advice.

The moment you paste a customer's details into an AI tool, or let an AI agent read your CRM, that data is being handled by another company's software. For most small businesses that is fine if you have thought it through, and costly if you have not.
This guide explains what to consider before customer data touches AI, drawing on the NIST AI Risk Management Framework and guidance from the US Federal Trade Commission (FTC). It is general information, not legal advice. Privacy rules vary by state and industry, so check with a qualified attorney about your own obligations.
Why customer data and AI need extra care
AI tools are different from ordinary software in a few ways that matter for privacy:
- You may be sending data to a third party every time you use a hosted tool.
- The vendor's terms decide what happens next: whether the data is stored, who can see it, and whether it can be used to improve their models.
- Outputs can reveal inputs. The OWASP project lists sensitive information disclosure as a risk, where a model reveals personal information, financial details, health records or confidential business data in its answers. See the OWASP description.
- Staff may use tools you have not approved, which is how data ends up somewhere nobody tracked.
What the FTC has said
In a January 2024 post, the FTC's Office of Technology addressed companies that offer AI models to others and the customers who use them. Three points are useful for any small business reading vendor terms or making its own promises:
- Promises count wherever they are made. If a company pledges not to use customer data to train or update its models, that commitment can create legal exposure when broken, whether it appears in marketing, terms of service or a marketplace listing.
- Quiet changes to terms are risky. Using data for a new purpose calls for clear notice and affirmative express consent, and burying a change in fine print or legalese may violate the law.
- Leaving things out can mislead. The FTC treats material omissions much like false statements.
Its summary line is worth remembering: there is no AI exemption from the laws on the books. The same applies to you: if your privacy policy or sales pages say something about data, your use of AI needs to match it.
The FTC's separate guide to protecting personal information sets out five principles that map neatly onto AI use: take stock of what you have, scale down what you collect and keep, lock it, pitch (dispose of) what you no longer need, and plan ahead for a breach.
A framework to keep it manageable
The NIST AI Risk Management Framework is voluntary, and NIST describes it as helping organizations manage AI risks to individuals, organizations and society. Its four functions, as described on NIST's core page, translate into practical privacy questions.
Govern: set the rules and the owner
Decide who is accountable for AI and data, what staff may paste into which tools, and what your vendors must promise. NIST says governance covers third-party software and data too, so vendor choices belong here.
Map: know what data goes where
NIST's Map function is about understanding a system's purpose, setting and potential impacts. For privacy, that means listing each AI tool, what data it sees, and what could go wrong if that data leaked or was misused.
Measure: test and track
Check how the tool behaves before and after rollout. Try it with fake customer records first. Look at what it stores and shows.
Manage: respond to problems
Prioritize the risks, reduce them, and prepare incident response. Know whom you would notify and how you would stop a tool quickly.
The vendor question list
Before you connect customer data to any AI feature, find the answers to these in the vendor's terms, privacy policy or support team.
| Question | What you want to hear | |---|---| | Is our data used to train your models? | No by default, or a clear opt-out | | How long do you keep our data? | A stated period, with deletion on request | | Who can access it? | Limited staff, with logging | | Where is it stored and is it encrypted? | Encryption in transit and at rest | | Do you use subcontractors? | Named, with equivalent obligations | | Can we export or delete everything? | Yes, with a described process | | What happens after a breach? | Prompt notice and a stated process | | Can the terms change without notice? | Notice given, with a way to object |
Keep a copy of the terms you agreed to, with the date. If they change, you will want to know what they said when you signed up.
A practical checklist
- List your AI tools, including free ones staff use on their own.
- Classify your data. Public, internal, customer personal, and regulated (health, payment, children's data).
- Decide what is never pasted: payment card numbers, government ID numbers, health details, passwords, anything under a confidentiality agreement.
- Mask before you share. Replace names and account numbers with placeholders when the task does not need them.
- Check the vendor answers in the table above and record them.
- Limit access. Give people and agents only the data their job needs.
- Turn off training use where the tool offers the setting.
- Update your privacy policy if AI changes how you collect, use or share customer data.
- Train your team on a one-page rule sheet and repeat it every few months.
- Plan for a breach, including who decides, who calls the lawyer and how customers are told.
- Dispose of data on schedule. Delete exports and chat histories you no longer need.
Regulated and sensitive data
Some data carries extra rules. Health information, payment card data, and information about children are examples where industry or state requirements may apply. Do not assume a general AI tool is suitable. Ask your attorney or compliance adviser before putting such data into any AI system, and prefer tools that state clearly how they handle it.
OWASP's mitigation advice for this risk includes scrubbing or masking sensitive content, applying least privilege, publishing clear retention and deletion policies, letting users opt out of training, and advising users not to enter sensitive information. It also cautions that instructions inside a prompt can be bypassed, so they should not be your only protection.
Common mistakes
- Assuming "private" means "not used for training." Read the terms.
- Pasting first, thinking later. One careless paste can expose a customer's record.
- Ignoring shadow tools. If staff use a personal account, your policy does not apply.
- Promising more than you do. Make sure your public statements match your practice.
- No deletion plan. Data you keep is data you can lose.
- Giving an agent broad access when a narrow slice would do. See AI agents and human approval.
For related routines, see our guides on record keeping and practical, safe AI uses.
Where WorkBOS fits
WorkBOS keeps customer records, projects, invoices, documents and chat in one system, which makes it simpler to see where customer data lives. Roles and permissions control who can see what, an audit log records activity, and AI agents act only with human approval. None of this replaces your own review of how data is handled. Look at the docs and pricing for details and ask questions before you move sensitive records.
Frequently asked questions
Q: Can I put customer information into a public AI chatbot? A: It is best avoided. If you need help with a task involving a customer, remove names and identifiers first, and read the tool's data terms.
Q: Does the NIST AI RMF have legal force? A: No. NIST describes it as voluntary. It is useful as a practical structure for thinking about risk, but it does not replace legal requirements.
Q: What does the FTC expect from small businesses using AI? A: The FTC says there is no AI exemption from existing laws. Keep your privacy promises, be clear when you change how data is used, and do not mislead customers by leaving out important facts.
Q: Do I need to update my privacy policy for AI? A: If AI changes how you collect, use, share or store customer data, your policy should say so accurately. Ask an attorney about your situation.
Q: What should I do if customer data is exposed? A: Follow a written response plan: contain the problem, investigate, and decide whom to notify, with legal advice on notification rules.
Sources are linked in the text above: NIST AI RMF, NIST AI RMF core, OWASP GenAI Security Project, and the FTC (January 2024 post on AI companies and privacy commitments; Protecting Personal Information guide).
Run it all from one place
Projects, clients, finance, people and AI employees in one system. Start free, no card needed.
Start free →