# AI Policy for Employees: Content and Rollout

Author: Marin Niehues, Niehues Consulting · Published: 2024-12-30 · Updated: 2026-11-16 · AI Law, Risk & Governance
Source: https://www.niehuesconsulting.com/blog/ai-policy-for-employees/

An AI policy for employees defines which AI tools may be used with which data and for what purpose, and it only takes effect with training and clear responsibilities. This article sets out the seven components with sample wording, places the legal framework and describes the rollout in four steps.

## Why does a company need an AI policy?

A company needs an AI policy as soon as employees use AI tools, because otherwise each individual decides which data leaves the company. This article provides an overview and does not constitute legal advice.

Employees have long been working with generative AI, often with their own accounts and before the company has a rule for it. They draft texts and summarize documents. They analyze spreadsheets with AI assistants. Official German statistics show that in 2024 one in five companies with at least ten employees already used AI. The benefit is real, and with it comes a question of control: which information leaves the company, and who checks the results?

A policy answers these questions uniformly across the organization. From a delivery perspective it is a steering instrument. It shortens decision paths, because a sales employee or a purchasing clerk need not ask questions before using an application. It also creates the basis on which business units adopt AI in their processes and executive management tracks the benefit through metrics.

**The legal framework in brief**
- AI literacy (Article 4 AI Act): providers and deployers of AI systems ensure that their staff have sufficient AI literacy. A policy with training is the most direct evidence.
- Transparency (Article 50 AI Act): labeling obligations apply to chatbots, generated images and video, and certain AI texts informing the public.
- Data protection (GDPR): personal data may only be processed on a legal basis and with providers covered by a data processing agreement.
- Co-determination (Germany, BetrVG): if the company provides AI tools capable of monitoring employee behavior or performance, the works council has a co-determination right under section 87(1) no. 6 BetrVG and must be informed of the planning under section 90.

There is no explicit legal obligation to have an AI policy. The AI Act does, however, require every company deploying AI to ensure sufficient AI literacy among its staff. A policy with training meets this requirement most directly and makes it demonstrable.

**An AI policy is a procedure with an owner and a review cycle that includes training. Only the procedure changes how an organization behaves.**

## What should an AI policy for employees include?

An AI policy includes seven components that cover every everyday decision, from the choice of tool to the reporting channels. It stays short enough for all employees to read and concrete enough to decide individual cases:

- Scope and definitions: who the policy applies to and how it relates to existing information security rules
- Approved tools: a list of applications the company provides or permits, with the permitted purpose
- Data classes: which information may go into which tool
- Responsibility for results: whoever uses AI owns the result, and in which cases a second review is mandatory
- Labeling and documentation: when AI support is disclosed to customers or partners, at least in the cases Article 50 of the AI Act prescribes
- Excluded applications: fields in which AI is not used without a separate review, such as decisions about individuals
- Reporting channels and consequences: whom employees contact in case of doubt or error

Where the policy touches data protection, copyright or employment law, drafting the individual rules belongs with the legal and data protection functions. The management task lies in asking the questions and translating the answers into a process.

## Is there a template for an AI policy?

A good template is an outline with sample wording that the company adapts to its own tools and data classes. Copying another company's policy verbatim fails on the details, because tool lists and data classes differ in every company. The following outline gives one sample clause per component as a starting point.

**Sample outline with example clauses**
- Scope: "This policy applies to all employees, trainees and contractors who use AI applications on behalf of the company."
- Tools: "The applications in Annex 1 are approved. Other applications are requested via the approval form before use."
- Data classes: "Confidential and personal data are entered only into applications explicitly approved for the purpose concerned."
- Responsibility: "Anyone using an AI output checks it for accuracy before passing it on. Contracts, figures in reports and statements to customers are additionally checked by a second specialist."
- Labeling: "Content that goes to customers or the public without human revision is labelled as AI-generated."
- Exclusions: "AI is not used for decisions on applications, appraisals or dismissals without approval from the AI committee."
- Reporting: "Errors and doubtful cases are reported to the AI mailbox. Reports made in good faith carry no disadvantage."

The tool list and examples per business unit belong in annexes. The core of the policy then stays stable, while the annexes can be updated quickly for each new tool.

**Questions to ask before every use of AI**
- Which data am I entering, and are they approved for this tool?
- Who checks the result before it leaves the team or the company?
- Is this use permitted for this purpose, or does it need approval?

## Data Classes Govern Everyday Use

Of all the components, data classification has the greatest practical effect. Employees can rarely judge how a provider processes inputs, but they know whether a document is public, internal or confidential. The policy therefore builds on this distinction and translates it into a rule per tool.

A proven pattern uses four levels. Public information may go into any approved tool. Internal information is permitted only in applications covered by a company contract. Confidential information and personal data require explicit approval for the specific use case. The fourth level covers information that never enters an external AI application. Where a company already classifies documents, the policy adopts that scheme. A second system only creates confusion.

The choice of tools determines how strict the rules feel in daily work. When employees have a vetted tool under a company contract, the incentive to use private accounts falls.

## How do you roll out an AI policy?

An AI policy is rolled out in four steps: survey use, agree the draft, enable employees, measure the effect. Only the rollout turns the document into a rule that applies.

The first step creates a picture of actual use. Short conversations with business units and a survey show which tools are used for which tasks. This picture determines which rules meet the greatest need.

In the second step, the draft is developed with the business units. Information security and data protection contribute, as do legal and HR. Where a works council exists, it belongs in this step: if the company provides AI tools that record usage data, it has a co-determination right under section 87(1) no. 6 BetrVG and must be informed of the planning under section 90. Where tools touch co-determination, the policy is often supplemented by a works agreement.

The third step is enablement. A short briefing per role with examples from the employee's own field works better than a general lecture. Employees practice classifying data and checking results on real cases. Attendance is documented, because it also evidences AI literacy under Article 4 of the AI Act.

In the fourth step, the company measures the effect. Metrics are the share of trained employees and the number of approved use cases, together with the number of reports and the duration of approval procedures. The figures show whether the policy is used or circumvented.

**Policy as a document and as a procedure**
Criterion | Document on the intranet | Procedure with an owner | With Niehues Consulting
Binding force | formal, unknown in daily work | known through training and examples | rollout with processes and enablement per role
New tools | used without approval | regulated approval procedure | approval route with deadlines and clear responsibility
Evidence of AI literacy | missing | training record per role | training per role, documented and linked to metrics
Maintenance | irregular | fixed review cycle | metrics and feedback from business units
Acceptance | low, circumvention likely | higher thanks to usable tools | prototypes in the real process, adoption by the business unit

## Responsibilities and Maintenance

Every policy needs one person who owns it. In practice this role sits with a manager from IT, information security or organization, with a clear mandate from executive management. A group for approvals and further development is added. How such bodies stay lean and how the policy fits into wider governance is shown in [Building AI governance](/blog/ai-governance-framework/).

Maintenance follows a fixed rhythm. The owner reviews the policy at least every six months and after every incident. Each change is announced briefly to all employees. Business units report where the rules slow work down, because a policy that brakes without need loses its authority. How the policy is anchored in overall AI management is described in the article [AI management: what executive leadership must steer](/blog/ai-management-what-executives-must-steer/).

## Niehues Consulting as Partner for Policy and Rollout

[Niehues Consulting](/) supports companies from the draft to adoption in daily work. In AI management, the picture of use and the requirements of the business units are established. In operationalisation and implementation, the policy with its approval procedure, the enablement per role and the AI applications that employees use instead of private accounts are created. In delivery management, Marin Niehues steers implementation as delivery manager, project manager or product owner. Individual legal questions are clarified by your company's specialists or your legal advisers.

The approach follows the Niehues Delivery Model with the phases Understand, Build and Implement. Three review questions matter most for the policy. In Understand: who decides, who delivers, who uses? In Implement: are users enabled and processes adapted? Also in Implement: which metric shows after 90 days that the solution is landing? Marin Niehues looks at organizations from a delivery perspective and has led projects from VC-funded scale-ups to DAX 40 groups in 20+ countries. The entry point is a free 30-minute initial consultation via [the contact page](/#kontakt).

**Recommendation for executive management**
Appoint an owner for the AI policy, give employees an approved tool and plan enablement before publication. The combination of rule, tool and training turns a document into an organization that uses AI reliably and can demonstrate its AI literacy.

**Sources**
- Federal Statistical Office of Germany (Destatis): Use of information and communication technologies in enterprises, 2024
- Regulation (EU) 2024/1689 laying down harmonized rules on artificial intelligence (AI Act), 2024

## Frequently asked questions

### Does every company need an AI policy?

Every company whose employees use AI tools needs an AI policy. There is no explicit obligation to have one. The AI Act does require sufficient AI literacy among staff, and a policy with training is the most direct way to demonstrate it.

### Is an AI policy mandatory under the EU AI Act?

The AI Act does not prescribe an AI policy, but it does require sufficient AI literacy among staff under Article 4. Companies usually meet this obligation with a policy and documented training per role. Labeling obligations under Article 50 apply in addition for certain content.

### What does AI literacy under Article 4 of the AI Act mean?

AI literacy means that employees working with AI systems understand their opportunities and risks and use them competently. The regulation requires measures that take prior knowledge and the context of use into account.

### Is there an AI policy template?

The best template is an outline of seven components with sample clauses, as shown in this article. Each company adapts the tool list and data classes to its own situation. Copying another policy verbatim produces rules that do not work in one's own daily practice.

### How long should an AI policy be?

An AI policy runs to two to four pages, supplemented by annexes with the tool list and examples per business unit. The annexes can be updated faster than the core. A short policy gets read, a long one gets filed.

### Can employees use ChatGPT with company data?

Employees may enter company data only into tools approved for the relevant data class. Public information is generally unproblematic, while internal information belongs only in applications under a company contract. Confidential and personal data require explicit approval for the purpose.

### What is the difference between an AI policy and a works agreement?

The AI policy is the employer's rule for using AI, while a works agreement is a contract between employer and works council. A works agreement becomes necessary when tools touch matters subject to co-determination, such as the analysis of usage data. The two documents refer to each other.

### Who should be involved in drafting the policy?

Business units and information security belong in the drafting, as do data protection, legal and HR. Where a works council exists, it is involved early so that participation questions are settled before rollout.

### What happens if employees breach the AI policy?

Breaches are handled like other breaches of company rules, graded by severity and intent. Errors that employees report themselves lead to correction and better training. Employment-law consequences in individual cases are for HR and legal advisers.
