Guide · template · checked 8 November 2026

AI use policy for a small company: a one-page template and how to roll it out

Staff are already using AI, with or without permission. A ban tends to push it out of sight; one clear page tells people what they can use, what must never go in, and who checks the result. Copy the template below and adapt it in an afternoon.

The short answer

In one paragraph

A small company’s AI use policy should fit on one page and answer five questions: which tools are approved, what data may go into each, who checks AI output before it reaches a client, what must be disclosed, and who to ask. Use a three-colour data rule, make it bind managers too, give the team a short training and keep a record of both. In the EU this also supports the Article 4 AI literacy duty. The template below is ready to adapt.

By Ian Luca Calcavechia, AI consultant, Kraków · Last updated 2026-11-13

Why one page matters now

  • Staff use AI anyway. In a June 2026 survey of 1,250 office workers at companies with $500 million or more in revenue, 66% had used AI tools they believed company policy did not allow, and 34% had entered customer data into public AI tools (PagerDuty and Wakefield Research). Large employers, so a pointer rather than a small-company measure.
  • A policy alone is not enough. In the same survey 86% said their employer had AI policies, and 81% believed leaders follow different AI rules from everyone else. The policy has to bind the owner too.
  • Sensitive data goes in. 58% of 500 US employees said they had entered sensitive data such as client records or financial data into AI tools (Anagram survey via HR Dive, August 2025).
  • The cost shows up in breaches. IBM’s 2025 report linked shadow AI to 20% of breaches and $670,000 of added average cost for organisations with high levels of it; 63% of breached organisations had no AI governance policy or were still writing one (FM Magazine on the IBM report).
  • Smaller teams often have nothing written. Among HR professionals on teams of 20 to 500 people, 80% used AI and 23% had a formal policy (HR Partner 2026 report; a vendor survey that does not state its sample size).

The rule that does most of the work: three colours for data

ColourWhat it coversWhere it can go
GreenPublic information: published prices, web pages, general questions with no namesAny approved AI tool
AmberInternal documents, drafts, supplier terms, client names without sensitive detailsOnly company accounts whose terms exclude training and that the owner has approved
RedHealth, financial and ID data, passwords and API keys, anything under an NDA, unpublished source code, staff recordsNever in a chat tool. Only inside a system built for it, with a signed data agreement

Why accounts matter: OpenAI’s help page says it may use content from its individual services to train models unless the user opts out, and that it does not use content from its business offerings, naming ChatGPT Team, ChatGPT Enterprise and the API (OpenAI, read 8 November 2026). Other vendors differ, so read the terms of each tool before approving it.

The template: copy and adapt

AI use policy · [Company name] · version 1, [date]
  1. Purpose. AI helps us work faster. People stay responsible for everything we send, publish or decide. This policy applies to everyone who works for us, including owners, managers and contractors.
  2. Approved tools. [Tool, plan, who may use it, for what]. Ask [name] before using anything else, including browser extensions and AI note-takers in meetings.
  3. Data. Green: public information, any approved tool. Amber: internal documents and client names, company accounts only. Red: health, financial and ID data, passwords, anything under NDA, unpublished code, staff records: never in a chat tool.
  4. Accounts. Work happens in company accounts, never in personal ones. Training on our data is switched off where the tool allows it.
  5. Check before it leaves. Every fact, figure, quote, citation and link in AI output is checked against its source by the person who sends it. That person’s name stays on the work.
  6. A person decides. AI does not decide hiring, firing, pay, credit, medical matters or a client complaint. It may prepare information; a named person makes the call.
  7. Tell people. Customers are told when they are talking to AI. AI-generated images or audio that could be mistaken for real are labelled. Clients are told about AI use when the contract asks for it.
  8. Ownership. Material we need to own, such as logos and key contracts, gets real human work and review. Ask [name] if unsure.
  9. Mistakes. If something Red went into a tool, or a wrong AI answer reached a client, tell [name] the same day. Reporting early is never punished.
  10. Training and review. Everyone joins a short AI training when they start and once a year. [Name] keeps the attendance list and the tool list, and reviews this page every six months.

A starting point, not legal advice. Have it checked against employment law, client contracts and the rules for your sector and country.

Three failures a policy prevents

Data · Samsung, 2023

Staff entered confidential source code and a full meeting into ChatGPT, and the company then barred such tools at work by internal memo, later calling the ban temporary (Fortune). Rule 3 of the template covers this.

Checking · Mata v. Avianca, 2023

Two New York lawyers filed a brief citing cases ChatGPT had made up and were fined $5,000. The judge said using a reliable AI tool for assistance is not inherently improper; failing to check is the problem (JD Supra). Rule 5.

Client work · Deloitte Australia, 2025

A AU$440,000 report for the Australian government contained fake citations and a fabricated court quote. Deloitte agreed to repay the final instalment, and the corrected version disclosed use of generative AI (The Register). Rules 5 and 7.

What the rules ask for

  • EU AI literacy. Companies that use AI systems must take measures to support the AI literacy of staff and others acting for them. After the 2026 Digital Omnibus they do not have to guarantee a level for each person, and the duty remains (KI-Campus). A policy, a short training and an attendance list are a practical record; dates are in the EU AI Act guide.
  • High-risk uses in the EU. AI that screens job applicants or scores credit falls under the heavier rules from 2 December 2027 (EU AI Act guide). Rule 6 keeps a small company clear of most of this.
  • Personal data in the EU. A vendor that processes personal data for you needs a contract that meets Article 28 of the GDPR (GDPR text). Ask for the vendor’s data processing agreement before approving a tool for Amber data.
  • Health data in the US. A covered entity that lets a cloud provider handle patient data without a business associate agreement violates HIPAA (HHS).
  • Ownership in the US. The Copyright Office said on 29 January 2025 that authors must be human, so purely AI-generated work appears to be ineligible, while human changes and arrangement can be protected (National Law Review).
  • A fuller framework if needed. NIST’s voluntary Generative AI Profile (AI 600-1, July 2024) is the reference for a company that outgrows one page.

General information, not legal advice.

Roll it out in two weeks

  1. Days 1–3: ask, without blame. A two-question form: which AI tools do you use, and for what? Honest answers need a promise that nobody gets in trouble for the past.
  2. Days 4–6: choose. List the tools, read each vendor’s terms on training and data, approve two or three, and move people to company accounts.
  3. Days 7–8: write. Fill in the template with real names and tools. Keep it to one page.
  4. Days 9–11: train for 60 minutes. Use the team’s own tasks and show one made-up citation being caught. Keep the attendance list.
  5. Days 12–14: put it to work. Pin the page where people work, add it to onboarding, and put the six-month review in a calendar.

Once the rules are clear, the useful next step is deciding what to automate: see what AI automation costs and how to hire an AI consultant.

Get it done

Start with a fixed package

AI use policy and team training

The Transformation plan package: roadmap, AI use policy, AI literacy plan and a 60-minute team training

Start on Fiverr →

AI quick scan

One team or process: where AI fits, quick wins and AI Act flags, in a short brief

Start on Fiverr →

All services and prices

Custom engagements for larger companies

Open →

Questions

Frequently asked

Does a small company need an AI use policy?

Yes, if anyone uses AI at work, because the risks come from everyday use: client data in a personal account or a made-up figure in a proposal. One page with a data rule and a checking rule is enough to start.

What should an AI use policy include?

Approved tools, what data may go into each, who checks output before it leaves the company, what must be disclosed to customers, which decisions stay with a person, how to report a mistake, and who owns the policy.

Should we ban ChatGPT at work?

A ban is hard to enforce: in a 2026 PagerDuty survey at large employers, 66% of office workers had used AI tools they believed were not allowed. Approving a company account and setting data rules gives more control than a ban.

Is an AI policy required by the EU AI Act?

The Act does not require a document by name. Article 4 requires companies using AI to take measures to support staff AI literacy, and a written policy with a short training is a practical way to show that.

Is the free version of ChatGPT safe for company data?

Not for internal or client data. OpenAI's help page says content from individual services may be used for training unless the user opts out, while its business offerings are excluded. Check the current terms of any tool before approving it.

Who can write an AI use policy and train my team?

Ian Luca Calcavechia, an AI consultant based in Kraków, includes an AI use policy, an AI literacy plan and a 60-minute team training in the Transformation plan package of his AI roadmap service on Fiverr.