
Copy a one-page AI acceptable use policy, apply the GDPR settings checklist, and assign a one-week rollout with owners and deliverables.
4 controls belong in every AI acceptable use policy: approved tools, data boundaries, human review, and an incident owner. Copy the one-page template below, adapt it to your organization, then roll it out in one week.

For the next step after ai policy template (2026), review AI for Recruitment Agencies: Keep Candidate and Client Data Boundaries Clear, and AI Vendor Security Questionnaire: Ask for Evidence Before Access.
Copy this template into your policy document, replace the bracketed fields, and assign a named policy owner.
Owner: [role or team] | Effective date: [date] | Review date: [date]
We use approved AI tools to assist employees with research, drafting, analysis, and other tasks that improve our work. AI supports employee judgment. It does not make a final business, hiring, customer, financial, safety, or legal decision on our behalf.
Use only the tools and account types listed in [approved tool register]. Sign in with your company account where available. Do not connect company files, email, or applications to an AI tool unless [role or team] has approved the connection and its permissions.
Employees may use approved tools to create a first draft, summarize public or approved material, organize ideas, analyze non-sensitive information, and improve routine work. Check tool-specific instructions before using AI for a new type of task.
Do not enter customer or prospect personal data, employee records, credentials, authentication codes, payment details, health or legal records, trade secrets, confidential contracts, unpublished financial results, security incidents, or information restricted by a client or partner. Do not paste data that identifies a person unless the tool and purpose have been approved for that data.
A qualified employee must check facts, calculations, citations, tone, privacy, and permissions before using AI output. Employees remain responsible for anything they send, publish, file, or use to make a decision. Do not present generated material as verified research unless you have checked the sources.
Do not use AI alone to rank or make decisions about a person, including hiring, lending, insurance eligibility, housing, health, discipline, or access to essential services. A manager and [risk or legal owner] must approve any proposed use in these areas before a pilot.
Follow [company disclosure rule] when a customer, regulator, or employee should know AI was used. Keep a record of the approved use case, tool, human reviewer, and source material when required by [records policy].
Stop use and notify [security or privacy contact] at [reporting channel] if you entered restricted data, receive an unsafe or discriminatory result, find a security issue, or suspect an AI-assisted mistake caused harm. Include the tool, time, data category, and action taken. Do not forward sensitive material in the incident report.
Complete [training] before using AI for company work. [Policy owner] reviews this policy every [review interval] and after a material tool, law, or risk change.
| Day | Decision | Deliverable |
|---|---|---|
| Monday | Name the sponsor, policy owner, and review team | Owner list and decision deadline |
| Tuesday | Inventory AI tools, account types, data, and current use | Approved-tool register and risk notes |
| Wednesday | Set allowed tasks, prohibited data, and review rules | Final one-page policy |
| Thursday | Configure accounts, access, retention, and escalation routes | Settings checklist and incident contact |
| Friday | Train one team with examples and collect questions | Signed acknowledgement and a 30-day review date |
Keep the first rollout small enough to support. The policy owner should be able to answer who can use each tool, which data is allowed, and where an employee reports an unsafe output or accidental disclosure.
GDPR does not certify a chatbot as compliant. Review the purpose, lawful basis, data minimization, transparency, processor terms, access, transfers, retention, and user rights for the workflow. The table summarizes the published vendor defaults and controls; record the exact edition and contract in your tool register.
| Tool | No-training default | Data residency or EU hosting | DPA availability | Retention controls |
|---|---|---|---|---|
| ChatGPT Business and Enterprise | Business data is not used to train by default | Eligible Enterprise customers can select EU storage at rest; Business does not have the same published residency option | OpenAI offers a DPA for Business and Enterprise | Workspace admins can control retention; deleted data is removed within 30 days, subject to legal exceptions |
| Claude Team and Enterprise | Commercial inputs and outputs are not used to train by default | Anthropic says data is stored in the US by default; routing through Europe is not EU data residency | DPA is incorporated into Anthropic Commercial Terms | Enterprise has custom retention controls; deleted conversations are removed from backend storage within 30 days under standard retention |
| Microsoft 365 Copilot | Prompts, responses, and Graph data are not used to train foundation models | Microsoft 365 commercial commitments include the EU Data Boundary | Covered by Microsoft's Products and Services DPA | Use Microsoft Purview and tenant retention policies for stored Copilot interactions and content |
| Gemini for Google Workspace | Content is not used to train outside the domain without permission | Data Regions policies can cover specified data and services in the US or Europe, depending on edition | Covered by Google's Cloud Data Processing Addendum | Admins set Gemini in Workspace prompt and response retention from 90 days to indefinite; Vault manages supported Workspace data |
| Krater Enterprise | Provider routing is limited to providers with a no-training commitment | EU infrastructure | DPA available | Data export and deletion are supported; a fixed retention period is not published |
Before approving a workflow, save the policy owner, account type, purpose, data category, location option, retention setting, and deletion process. A vendor's default does not replace your organization's decision about why the data is processed or how long the business needs it.
Employees need examples, not a long list of abstract risks. Show one allowed prompt using public information, one prompt that must be rejected because it contains personal data, and one example of a human review. Make the policy easy to find in onboarding and repeat training when tools or account settings change.
Measure adoption by completed work and errors caught, not by prompt count. Review the tool register quarterly, assign an owner to access and retention, and update the policy when a team proposes a new data source or high-impact use.
Create documents, slides, sheets, research, images, and video across 500+ AI models in one workspace. A team can create a Persona for a recurring task, keep role instructions reusable, and use a Teams workspace for shared work.
For rollout, create one Persona for policy questions, one for approved training drafts, and one for internal FAQs. Keep the approved-tool register as the source of truth, and require a person to review any answer that changes an employee's obligations. Adam can help map the workflow and rollout to the team's size and usage in a free AI consultation. Free, no strings attached.
For a free AI consultation, Adam maps what is best for your team size and usage. Free, no strings attached.
Thirty minutes with Adam Johansson, an AI expert who maps out what actually fits your business: the right tools, plan and setup for your team size and usage. Free, no strings attached.
Free for any business, any size. We reply within one business day.

Adam Johansson
AI Advisor, Krater
A good AI acceptable use policy is short, specific, and connected to real settings. Copy the template, name an owner, test the vendor controls, and train staff with examples. Use the one-week rollout to make the rules operational, then review the approved tools and data boundaries on a schedule.
List approved tools, allowed tasks, prohibited data, human review, disclosure rules, an incident route, a policy owner, training, and a review date. Add stricter rules for high-impact decisions and regulated information.
Yes. Replace every bracketed field, name the tools and roles that apply to your organization, and have the appropriate privacy, security, HR, and legal owners review it before adoption.
No. You still need to define the purpose and lawful basis, minimize data, provide required information, evaluate the processor terms and location, set retention, and support data-subject rights.
Name an owner, inventory tools, approve allowed use and prohibited data, configure accounts, and train the first team. The five-day plan above gives each step a deliverable.
Do not delegate high-impact decisions to a general AI tool. Use a qualified person and the organization's approved process for hiring, housing, insurance, health, lending, discipline, and access to essential services.
Sources: General Data Protection Regulation, European Commission data protection, NIST AI Risk Management Framework, Responsible AI Institute policy template, SHRM AI policy guidance, ANB Advisory policy template, Krater Enterprise. Regulatory references checked November 2026. Additional official sources checked November 2026: OpenAI enterprise privacy; OpenAI business data and residency; OpenAI Data Processing Addendum; Anthropic commercial training policy; Anthropic data location; Anthropic DPA; Anthropic retention controls; Microsoft 365 Copilot privacy; Microsoft Copilot enterprise data protection; Google Workspace Gemini privacy; Google Workspace Data Regions; Google Workspace privacy and CDPA; Krater Enterprise.