Home › AI for small firms › Data, privacy and the rules
Data, privacy and the rules
AI policy for a small UK firm: a one-page template to copy, with the rule behind each line
A ten-line AI policy a small UK firm can copy today: which tools, what never goes in, who checks, what to do when it is wrong, and the rule behind each line.
A small firm's AI policy fits on one page: the ten lines below can be copied today, and each rests on a rule or a vendor term read on 1 October 2026. None of the ICO pages we read requires a document called an "AI policy", but most of the lines state things the law does require, among them a written contract with the supplier, a person involved in any significant decision about someone, no fake reviews, and a notifiable data breach reported within 72 hours.
The table after the policy says which lines are law and which are good sense.
Free tool
AI rules notice to staff maker
Fill in the tool the firm pays for, who checks output and who to tell when something goes wrong. The notice sets out the house rules for AI assistants, as a short email, a cover note for the full policy, or a reminder.
- Attach the signed one-page policy to the second and third versions.
- Check the tool named is one the firm has a data processing agreement for, and that the privacy notice names that supplier.
- Send the reminder to everyone and keep names out of it; deal with the person concerned separately.
- Where personal data went into a personal account, the 72 hours to decide whether to report runs from when the firm became aware of it, not from the date of the reminder.
- Keep a dated copy of each notice with the policy.
The rule this rests on: ICO, written contract with every processor, data minimisation and breach reports within 72 hours; UK GDPR Articles 22A to 22D; Digital Markets, Competition and Consumers Act 2024, Schedule 20 (fake reviews). A draft to check and adapt, not legal advice.
Works in your browser. Nothing you type leaves this device.
The policy
Copy it, fill in the brackets, sign it and date it.
[Firm name]: how we use AI
- Tools. We use these AI tools, on business accounts the firm pays for: [list]. Nobody uses a personal account for work.
- Before a tool is added. [Name] reads its data processing agreement, where it holds data and what it says about training, and records whether a data protection impact assessment is needed.
- What never goes in. Bank and card details, passwords, health information and anything about a child go into no AI tool.
- As little as the job needs. Other customer details go only into a tool on the list, and only the details the task needs.
- Checking. The person who sends or uses the output checks it first: every figure, date, name and statement of law against its source.
- Decisions about people. No tool decides alone whom we hire, dismiss, refuse or charge more. A named person looks at each such decision and can change it before it takes effect.
- Being straight with customers. Our privacy notice names our AI suppliers. Nothing we send or publish pretends a machine is a person, and we never post a customer review that a customer did not write.
- Feedback buttons. Nobody presses thumbs-up or thumbs-down on a chat that contains customer details.
- When it goes wrong. Tell [name] the same day. If personal data has gone where it should not, [name] decides whether the ICO must be told, which has to happen within 72 hours.
- Keeping it current. [Name] reads this policy again every six months and whenever a tool or its terms change.
Signed: [owner], [date]
The rule behind each line
| Line | What it rests on | Law or good sense |
|---|---|---|
| 1 | The ICO says every time a controller uses a processor there must be a written contract. OpenAI lists its data processing addendum for ChatGPT Business, Enterprise and the API. Anthropic's sits inside its commercial terms. Neither page lists one for a personal plan | Law |
| 2 | The ICO lists AI as innovative technology. A data protection impact assessment is required when that is combined with another risk criterion, and a firm that decides one is not needed still documents why | Law |
| 3 | The ICO says special category data, which includes health data, needs a lawful basis and an additional condition. Sensitive data combined with AI is one of the pairings that requires an assessment. The rest of the line is caution | Part law, part good sense |
| 4 | The data minimisation principle: personal data must be limited to what is necessary for the purpose | Law |
| 5 | The ASA said on 11 June 2026 that an advertiser remains responsible for any AI output used in its ads. Section 226 of the Digital Markets, Competition and Consumers Act 2024 makes false or misleading information to a consumer a misleading action | Law for adverts and consumer information, good sense elsewhere |
| 6 | Articles 22A to 22D of UK GDPR, fully in force since 5 February 2026. A significant decision taken with no meaningful human involvement triggers safeguards. The ICO's draft guidance of 31 March 2026 gives automatic refusal of online credit as an example, and says spot checks are not enough | Law |
| 7 | The ICO says a privacy notice must give the recipients of the data, including anyone processing it for the firm. Schedule 20 of the 2024 Act bans fake consumer reviews, in force since 6 April 2025, and says writing "includes creating by any means" | Law |
| 8 | OpenAI and Anthropic both say content sent as feedback may be used for training, even on a business plan. Anthropic says the owner of a Team plan can switch the buttons off for everyone | Vendor term |
| 9 | The ICO says a notifiable breach must be reported "not later than 72 hours after becoming aware of it". A data protection complaint from a customer must be acknowledged within 30 days | Law |
| 10 | The ICO calls a data protection impact assessment a live document. The nine dated vendor pages read for these guides had been updated between 8 January and 24 September 2026 | Good sense |
On line 6, the person named has to do real work: the draft guidance says the involvement must be "active and not just a token gesture". Our guide to telling customers you used AI covers what a firm must say when a tool does decide alone.
Line 2 has the most reading behind it. The five checks are in is it legal to put customer data into ChatGPT, Claude or Copilot.
Add your trade's rules
The ten lines are the floor. Several of our sector guides carry a policy written for one trade:
- Solicitors: the SRA's warning notice of 17 August 2026.
- Recruiters: a policy for candidate data, built on the ICO's six questions.
- Accountants: the client consent rule in the tax bodies' guidance of 19 January 2026, and a paragraph for the engagement letter.
- Letting agents: a three-line house rule that keeps prescribed notices on the government form.
- Clinics: the extra footing that patient records need before line 3 is relaxed.
- Financial advisers: six conditions before a tool reads a client file.
Where the sector rule is stricter, it wins.
Getting staff to read it
- Keep it to one page. Explanations go in a separate note.
- Go through it at a staff meeting and have each person sign the same copy.
- Pin it where the work is done.
- When a tool is added or dropped, reissue the page with a new date.
- Give line 5 a method. The five-step check for AI mistakes is one.
What the policy is not
It does not replace the documents it points to. The contract with the supplier, the privacy notice and any data protection impact assessment are separate; the policy records that they exist and who holds them.
Much of the ICO guidance behind lines 1 to 4 carries a banner saying it is under review after the Data (Use and Access) Act 2025, and the guidance behind line 6 is still a draft. Line 10 exists for that reason.
Overlyft
is an AI agency that sets up AI office assistants for ten trades: they are set up in the firm's own office for a fixed price agreed after an audit, the firm pays Anthropic directly for its own Claude plan, and a person approves anything that goes to a customer.
This page is general information, not legal advice.
Short answers from this page
Does a small UK firm need an AI policy?
No ICO page read on 1 October 2026 requires a document by that name, but a one-page policy records what the law does require: a written supplier contract, a person involved in significant decisions (UK GDPR Articles 22A to 22D, in force 5 February 2026), no fake reviews, and breach reports within 72 hours.
Source: ICO guidance; legislation.gov.uk
Where this comes from: ICO, "When is a contract needed and why is it important?", "When do we need to do a DPIA?", "Guidance on AI and data protection" (updated 15 March 2023), "Data minimisation", "What privacy information should we provide?", "Personal data breaches: a guide", "How to deal with data protection complaints" and the draft guidance "Automated decision-making, including profiling" (updated 31 March 2026); UK GDPR Articles 22A and 22C and the Digital Markets, Competition and Consumers Act 2024, sections 225 and 226 and Schedule 20, on legislation.gov.uk; ASA, "AI and Deepfakes: Four Things Advertisers Need to Know Before They Hit Run" (11 June 2026); OpenAI "Enterprise privacy" (8 January 2026); Anthropic privacy centre (19 August 2026, 16 March 2026); all read in a browser on 2026-10-01. Not legal advice.
AEO Lookup editorial team
We write practical guides for people who run small UK firms, and we check every fact against its source on the date shown. Read how we review, or tell us if something has changed.



