
An AI policy template, and the AI literacy duty behind it
Most AI policy templates you can download are either a page of slogans or a twenty-page code of conduct nobody reads. What a policy actually has to do is short: say which tools are allowed, what never goes into them, who is responsible for the output and who explains all of this to whom. Since February 2025 that last part is also a legal duty in the EU. This is the template we work from, chapter by chapter.
A usable AI policy fits on one page, with a tool register as an annex. The register changes monthly; the policy should not.
Article 4 of the EU AI Act requires organisations using AI to ensure, to their best extent, a sufficient level of AI literacy among the people working with it on their behalf. No exam, no certificate, but you must be able to show what you did.
Sufficient depends on the role. Someone rewriting an email needs different explanation from someone using AI in recruitment. Whether your policy meets the law is for a lawyer to judge; that is not something we do.
What the policy is for
An AI policy is not there to stop people using AI. In most organisations they already do, often on a personal account, and a ban only moves that use to a phone you cannot see. The policy is there so that people know which tools they may use and how, and so that you can explain afterwards what was agreed.
Which tool to choose, what business terms matter and why permissions are the real problem is covered in our article on AI in the workplace. This piece is about the document itself and the AI literacy duty that sits behind it.
One test before you start: if an employee cannot answer “may I use the tool I have open right now, and with this data?” after reading your policy, it is not finished.
An AI policy template: the chapters
The core fits on one page. What changes often (the tools) goes into an annex you update without re-approving the policy. These are the chapters:
- Scope. Who it applies to, including contractors and temporary staff, and which use: work accounts, but also a personal account used for work.
- Approved tools. A register listing per tool the name, the business tier, the owner, which data classes are allowed and when it was last reviewed. This is the annex.
- Data. What never goes in, linked to the classification you already use for confidential information, so you do not build a second system next to it.
- Responsibility. Whoever sends it is the author. And no decision about people (applications, appraisals, case files) comes out of a model without a person looking at it.
- What never happens without review. Since February 2025 the AI Act prohibits certain uses, including emotion recognition in the workplace, and it places heavy requirements on uses such as AI in recruitment. State that such a use goes past privacy and legal first.
- Requesting a new tool. To whom, with which questions, and how quickly there is an answer. A policy without this route gets bypassed.
- Reporting. What someone does after pasting customer data into a tool by mistake. This links to your data breach procedure.
- AI literacy. Which group gets which explanation, when, and how it is recorded.
- Owner and review. Who maintains the policy and when it is looked at again: periodically, and whenever a new tool is added.
What the AI Act means by AI literacy
The EU AI Act entered into force on 1 August 2024 and applies in stages. Article 4, on AI literacy, has applied since 2 February 2025. In short: providers and deployers of AI systems take measures to ensure, to their best extent, a sufficient level of AI literacy among their staff and other persons dealing with those systems on their behalf.
- A deployer is any organisation using AI under its own authority. You do not have to build anything; business use of an AI assistant is enough.
- Sufficient is not a fixed level. The text itself says to take account of people’s technical knowledge, experience and training, the context of use and the persons the system is used on.
- There is no exam. The regulation prescribes no certificate, number of hours or accredited course.
The regulation sets no specific fine for this article; supervision and penalties are left to Member States. There are also European proposals to amend parts of the AI Act, this article included, so follow the current position with your lawyer. If you operate only in the UK, the AI Act does not apply to you directly; once you use AI in the EU or its output is used there, it can.
AI literacy by role, not one course for everyone
The temptation is to give everyone the same e-learning and tick the box. That is evidence of something, but it misses the point of Article 4, which asks for explanation that fits what people do. What works in practice is three layers:
- Everyone who uses AI. What a language model is and why it can be convincingly wrong, which tools are approved and why the free ones are not, what never goes in, that the output is a draft you sign for, and where to report a question or a mistake. An hour, with examples from their own work.
- Teams that rely on it heavily or handle sensitive data. HR, customer service, finance, legal, developers. They get the risks of their own use: bias in assessments, invented sources in a legal draft, customer data in a prompt, code under a confidentiality clause.
- Those who decide or administer. Management, IT, procurement and privacy. They need to know what business terms say and when a use moves into a heavier risk category.
Do not forget contractors and new starters: the first work on your behalf too, the second arrive after the one-off session. For the practical side of a programme like this, see training and adoption.
How you show it is in place
Because there is no certificate, your own records are the evidence. Keep:
- The version of the policy that applied, with its date and who approved it.
- Per group, the content of the explanation and why it fits that group’s use.
- Who received which explanation and when, contractors and new starters included.
- The tool register with the date of the last review per tool.
Tie the basic explanation to onboarding so it does not depend on an annual campaign, and tie it to the register: when a tool with a new kind of use is added, the group using it gets an extra session. If you have works councils or employee representatives in EU countries, involve them early; tools that can monitor staff may require their consent or consultation.
Where it goes wrong
- A ban with no alternative. It moves the use to personal devices, where you see nothing and can explain nothing.
- A policy with no tool list. “Use AI responsibly” is not an agreement. People want to know whether the tool they have open is allowed.
- One session and done. The tools change faster than the policy; without review it is out of date within months.
- IT as the only owner. Data rules belong with privacy, appraisal rules with HR, the risk question with legal. IT supplies the technology and the register.
- A template copied without editing. A downloaded AI policy template names tools you do not use and misses the ones you do. Start from your own inventory.
Where our work stops
What we do. We map which AI tools run on your devices and in your tenant, set up business licences, block consumer versions on work accounts and tighten overly broad permissions before an assistant can reach them. For the policy we supply the technical part: the tool register, the settings per tool and the data rules as far as they can be enforced technically. We give practical explanation of the tools we manage and provide the records of it for your files.
What we do not do. We are not lawyers. We do not determine whether you comply with the AI Act, whether a use is high-risk or prohibited, and we do not sign off your AI policy legally. We do not issue a recognised AI literacy certificate either, and the regulation does not ask for one. Have the policy and the risk classification reviewed by a lawyer or privacy adviser.
Questions we get about this
The five that come up most often.
What should an AI policy include?
Who it applies to, which tools are approved, what data never goes into them, who stays responsible for the output and which uses never happen without review. Add how to request a new tool, how to report a mistake, who receives which AI literacy explanation and who maintains the policy. The core fits on one page, with a tool register as an annex.
Is AI literacy a legal requirement?
In the EU, yes, for organisations that provide or use AI systems. Article 4 of the AI Act has applied since 2 February 2025 and asks you to ensure, to your best extent, a sufficient level of AI literacy among staff and others working with AI on your behalf. European proposals to amend parts of the AI Act are under discussion, so check the current position with a lawyer.
Do employees need an AI literacy certificate?
No. The regulation prescribes no exam, certificate or number of hours. What you do need is your own record of which group received which explanation, why it fits their use, and when. Those records are your evidence.
Can we just use a free AI policy template?
As a starting point, yes, but not unedited. A downloaded template names tools you do not use and misses the ones you do, and it has no owner in your organisation. Start from an inventory of what actually runs, then fill in the chapters.
Can you tell us whether our AI use complies with the AI Act?
No. We are not lawyers and do not determine compliance or which risk category a use falls into. We map what AI is running, supply the technical part of the policy and configure the tools so the rules can be enforced. The legal review is for a lawyer or privacy adviser.
Further reading
What this policy connects to.
Shall we start by mapping the AI you already run?
That inventory is the basis of your tool register and of the question who needs which explanation. The legal review we leave to your lawyer.
Practical IT knowledge in your inbox
New guides on management, security and the workplace, written by the people doing the work. No sales talk, and you can unsubscribe in one click.