To keep your AI work private and secure, sort your information into levels of sensitivity, decide which levels may ever go into an AI tool, read each tool’s data terms before you use it, limit what the tool can access, remove identifying details before pasting anything, protect your accounts with strong authentication, and write down a short policy so you and anyone who helps you follow the same rules. Most AI privacy problems come from habit, not hacking: someone pastes a confidential document into a tool without thinking. A few clear rules prevent most of the risk.
Why this matters for a small business
When you use AI tools, you send your information to someone else’s systems. For a solo founder, that information often includes client details, financial records, business plans, and personal data. A leak can mean lost clients, legal trouble, or damage to your reputation that a small business cannot easily absorb.
| Risk | Example |
|---|---|
| Client confidentiality breach | Pasting a client’s strategy into a tool whose terms allow retention |
| Personal data exposure | Uploading customer lists with names and emails |
| Credential theft | Pasting passwords or access keys into a prompt |
| Account takeover | Weak login on a tool connected to your email |
| Contract violation | Breaking a client agreement that bans third-party tools |
| Regulatory trouble | Mishandling personal data covered by privacy laws |
The point is not to avoid AI. The tools are valuable. The point is to use them with the same care you would give any outside contractor who sees your work.
If you are building a business that needs fewer of your hours, the ExpatBuildr Optimize guide shows how to set up systems like this safely at the Optimize roadmap, and you can book a 1-on-1 call to review how your current tools handle your data.
Classify your information
You cannot protect everything equally. Sort your information into levels, and set a rule for each.
| Level | Examples | Rule for AI Tools |
|---|---|---|
| Public | Published articles, public pricing | Fine to use freely |
| Internal | Process notes, drafts, general plans | Use approved tools; avoid unnecessary detail |
| Confidential | Client work, contracts, financial records | Only with vetted tools and permission; anonymize where possible |
| Restricted | Passwords, keys, ID numbers, health or banking details | Never put into AI tools |
Write this table down, adapt it to your business, and keep it somewhere visible. The simple act of deciding in advance prevents most mistakes made in a hurry.
Read the terms before you trust a tool
The terms and privacy policies of AI tools vary widely, and they can differ between free and paid plans. Before using a tool for anything beyond public information, find the answers to these questions.
| Question | Why It Matters |
|---|---|
| Is my content used to train the provider’s models? | Your information could influence future outputs |
| How long is my data stored? | Longer storage increases exposure |
| Who can access it, including staff and subprocessors? | Wider access means more risk |
| Where is it stored? | Location affects which laws apply |
| Can I delete my data and my account? | You need an exit |
| Is there a business plan with stronger protections? | Terms are often better on paid business tiers |
| What security certifications or practices does it describe? | Signals maturity |
| What happens in a breach? | Notification commitments |
If the answers are unclear or unfavorable, treat the tool as suitable for public and low-sensitivity information only. Re-check terms periodically, because they change.
Check your own obligations
Your clients and the law may restrict what you can do.
- Read your client contracts for confidentiality clauses and rules on third-party tools.
- Ask before using AI on a client’s sensitive material if the contract is silent.
- Know the privacy laws that apply to the people whose data you handle, which depends on where they live, not only where you do.
- Keep records of what you shared and with which tool.
- Tell clients how you use AI when it matters to them. Transparency builds trust, and some clients require it.
When the stakes are high, such as health, legal, or financial data, get professional advice before using AI tools at all.
Reduce what you share
The safest data is the data you never send. Habits that help:
- Anonymize. Replace names, company names, and identifiers with placeholders.
- Summarize first. Send a short description instead of the full document.
- Remove numbers that are not needed for the task.
- Share excerpts, not whole files.
- Use fictional examples when testing a prompt.
- Strip metadata from documents before uploading.
A simple routine before pasting is to ask: if this appeared on a public page tomorrow, would I be comfortable? If not, change it or leave it out. The guide to prompt engineering for founders shows how to give tools useful context without oversharing.
Control access and connections
Many AI tools connect to your email, files, calendar, or other apps. Each connection is a door.
| Control | Practice |
|---|---|
| Minimum permissions | Grant read access only when that is enough |
| Separate accounts | Use a dedicated account for experiments |
| Review connections | Remove ones you no longer use |
| Limit scope | Connect a single folder, not your entire drive |
| Check what actions are allowed | Avoid tools that can send or delete without review |
| Watch for third-party add-ons | Extensions and plugins can carry risk |
Be cautious with tools that ask for broad access, particularly those that can read your whole inbox. The guide to building an AI chief of staff for your business discusses how to grant access in a limited, reviewable way.
Protect your accounts
An attacker who takes over your AI account can read your history and any connected systems.
- Use a password manager and long, unique passwords.
- Turn on two-step verification everywhere it is offered, preferably with an authenticator app or security key rather than text messages.
- Do not share logins. Use proper team access if you have helpers.
- Log out on shared or public devices.
- Keep devices updated and protected.
- Watch for phishing. Be wary of emails or messages asking you to log in or share a code.
- Review account activity for unfamiliar sessions.
Never type passwords, access keys, or recovery codes into a prompt. Treat anything you type into an AI tool as something that may be stored.
Beware of prompt injection and bad outputs
AI tools that read outside content, such as web pages, emails, or documents, can be tricked by hidden instructions inside that content. A malicious message might tell the tool to reveal data or take an action you did not intend.
| Practice | Why |
|---|---|
| Do not give tools authority to act on untrusted content without review | Prevents hijacked actions |
| Keep a human approval step for sending, deleting, or paying | Catches mistakes |
| Be cautious with links and files in AI output | They may be unsafe |
| Verify claims in outputs before acting | Outputs can be wrong or manipulated |
The more power a tool has, the more important review becomes. The guide to auditing and improving your AI stack over 90 days explains how to schedule regular checks of what each tool can do.
Write a one-page AI use policy
A short written policy turns good intentions into habits. It should be one page and cover:
| Section | Content |
|---|---|
| Approved tools | The list, and what each may be used for |
| Data levels | The table above, with rules |
| Never share | Passwords, keys, IDs, restricted data |
| Review rules | What must be checked by a human before it leaves |
| Account rules | Password manager and two-step verification |
| Client rules | How you handle client information and disclosure |
| Incident steps | What to do if something goes wrong |
Share it with anyone who helps you, and ask them to confirm they have read it. Review it twice a year.
If something goes wrong
Have a plan before you need one.
- Stop the exposure. Change passwords, revoke access, disconnect the tool.
- Assess what was involved. What data, whose, and for how long.
- Contain it. Ask the provider to delete the data if possible.
- Check your obligations. Contracts and laws may require you to notify clients or authorities within a set time.
- Tell affected clients promptly and honestly.
- Fix the cause, and update your policy.
Acting quickly and openly usually preserves more trust than hoping nobody notices.
Common mistakes
- Pasting confidential data without thinking. Classify first.
- Skipping the terms. Read data handling rules.
- Using free tools for sensitive work. Check plan differences.
- Over-broad connections. Grant minimum access.
- Weak account security. Use a password manager and two-step verification.
- Trusting AI output blindly. Verify and review.
- No written policy. Habits need documentation.
- No incident plan. Prepare in advance.
A starter plan
| Day | Task |
|---|---|
| 1 | Write your four data levels and the rule for each |
| 2 | List every AI tool you use and read each one’s data terms |
| 3 | Review client contracts for confidentiality and tool restrictions |
| 4 | Remove unneeded connections and tighten permissions |
| 5 | Set up a password manager and two-step verification on every tool |
| 6 | Write your one-page AI use policy |
| 7 | Write your incident steps and set a six-month review date |
For the wider approach, see the guide to what AI arbitrage is. For the full set of guides, see the AI Arbitrage hub, and the AI arbitrage resource links page collects the tools mentioned across the pillar.
Frequently asked questions
Q: What should I never paste into an AI tool? A: Passwords, payment card numbers, government ID numbers, and confidential client data unless you have checked the tool terms and have permission. When in doubt, remove or mask identifying details first.
Q: How do I check whether an AI tool is safe to use with business data? A: Read its privacy and data retention terms, check whether your inputs are used for training and whether you can turn that off, and look at its security documentation before sharing anything sensitive.
Q: What should I do if sensitive data was shared by mistake? A: Stop using that session, delete the data where the tool allows it, change any exposed credentials, and tell affected clients if your contract or local rules require it. Check the rules that apply to you.
References
- National Institute of Standards and Technology, AI Risk Management Framework and Cybersecurity Framework
- Federal Trade Commission, Protecting Personal Information: A Guide for Business
- European Commission, Guidance on Data Protection for Small Businesses
- Cybersecurity and Infrastructure Security Agency, Small Business Cybersecurity Resources