A free AI policy template for companies, and the tools that enforce it

Contents

1. Why companies need a written AI policy now
2. The AI policy template
3. How we evaluated the enforcement tools
4. The six tools, ranked
5. The tools compared
6. The criteria
7. What the evidence says about enforcement
8. Rolling out the policy: the first 60 days
9. Questions companies ask
10. Sources

In brief

A usable company AI policy fits on a few pages: who owns it, which tools are approved, which data never goes into AI, which AI actions run automatically, which need a person’s approval and which are never allowed, plus training and an incident route. The template below covers all of it. Enforcement is where tools differ: for AI agents that act on company records, OpenLazarus is our top pick; for data going into chat assistants, start with the suite you already run.

• AI agents acting on company records: OpenLazarus. The company decides per AI teammate what runs automatically and what waits for approval, and its industry pages state that every action is logged.
• Microsoft 365 companies: Microsoft Purview. Data loss prevention for Copilot and agents, prompt and response records, and visibility of third-party AI sites.
• Google Workspace companies: Google Workspace admin controls. Published data terms for Gemini, admin on/off controls, audit logs and DLP on generated output.
• Regulated firms building an AI inventory: Credo AI. Ready policy packs for the EU AI Act, NIST AI RMF and ISO/IEC 42001, with agent and vendor registries.
• Firms that build their own models: IBM watsonx.governance. Lifecycle risk management and monitoring, with published starting prices and a self-managed option.

Why companies need a written AI policy now

AI arrived in most companies before any rule about it did. In Microsoft's 2024 Work Trend Index, which surveyed 31,000 knowledge workers in 31 markets, 75% said they used AI at work and 78% of those users brought their own AI tools rather than waiting for company-issued ones⁠1AI at work is here. Now comes the hard part (2024 Work Trend Index)Microsoft WorkLab · 8 May 2024. The same survey found that 60% of leaders worried their organisation lacked a plan to implement AI⁠1AI at work is here. Now comes the hard part (2024 Work Trend Index)Microsoft WorkLab · 8 May 2024.

The cost of that gap is now measurable. In IBM's 2025 Cost of a Data Breach study, conducted by the Ponemon Institute across 600 organisations, 13% reported breaches of AI models or applications. Of those, 97% had no AI access controls in place, and 63% of breached organisations either had no AI governance policy or were still developing one. Organisations with high levels of unsanctioned "shadow AI" saw breach costs on average $670,000 higher than those with little or none⁠2IBM Report: 13% of organizations reported breaches of AI models or applications, 97% of which reported lacking proper AI access controlsIBM Newsroom · 30 Jul 2025.

In the European Union there is also a legal floor. Article 4 of the AI Act has required providers and deployers of AI systems to take AI literacy measures for their staff since 2 February 2025, and national authorities supervise and enforce that duty from 2 August 2026⁠3AI Literacy: Questions & AnswersEuropean Commission · 20 Oct 2026. The work itself has also changed. AI tools no longer only draft text; agents now send email, update records and prepare payments. A policy written for chat assistants says nothing about which of those actions an agent may take on its own.

The AI policy template

The template below is complete and free to use. Replace every field in square brackets, delete clauses that do not apply and keep the numbering so that training material and audits can refer to it. Appendix A (approved tools) and Appendix B (the action table in clause 7) are designed to change more often than the policy itself. A clause-by-clause commentary, with variations for small and regulated companies, is in our guide to adapting the template.

1. Purpose

• 1.1 This policy sets out how [Company name] ("the Company") uses artificial intelligence (AI) so that it benefits our customers, staff and business while protecting confidential information, personal data and the Company's obligations.
• 1.2 It applies the principle that people remain accountable for work done with AI, whoever or whatever produced it.

2. Scope

• 2.1 This policy applies to all employees, contractors, temporary staff and anyone else working under the Company's direction ("staff").
• 2.2 It covers every AI tool used for Company work, whether provided by the Company or not, including chat assistants, AI features inside other software, AI agents that act in Company systems, and AI used by suppliers on the Company's behalf.
• 2.3 It applies on Company and personal devices whenever Company information is involved.

3. Definitions

• 3.1 AI tool: software that generates text, images, audio, code, predictions or decisions from input data, including AI features embedded in other products.
• 3.2 AI agent: an AI tool that can take actions in Company systems, such as sending messages, creating or changing records, or preparing payments.
• 3.3 Approved tool: an AI tool listed in Appendix A, used through a Company account under the conditions listed there.
• 3.4 Confidential information: [define, or refer to the Company's information classification policy].
• 3.5 Personal data: any information relating to an identified or identifiable person, as defined in [applicable data protection law, e.g. the GDPR].
• 3.6 Tool owner: the manager accountable for how an approved tool or agent is configured and used.

4. Roles and responsibilities

• 4.1 [Executive owner, e.g. Chief Operating Officer] owns this policy, approves changes and receives a report on AI use at least [twice a year].
• 4.2 [AI lead, e.g. Head of IT] maintains Appendix A, runs the approval process in clause 12, organises training under clause 13 and keeps the records in clause 15.
• 4.3 Each tool owner configures their tool or agent within this policy, reviews its activity at least [monthly] and reports incidents.
• 4.4 [Data protection officer or privacy lead] advises on any use involving personal data.
• 4.5 Every member of staff is responsible for following this policy and for the accuracy of work they submit, including work produced with AI.

5. Approved AI tools

• 5.1 Staff may use only the tools listed in Appendix A, for the purposes and data classes listed there, through Company-provided accounts.
• 5.2 Personal or free accounts of AI tools must not be used for Company work involving anything other than public information.
• 5.3 AI features that appear in software the Company already uses are not approved until listed in Appendix A.
• 5.4 Requests for new tools follow clause 12.

6. Data rules

• 6.1 Never enter into any AI tool: passwords, keys and access credentials; payment card or bank details; special category personal data (health, biometric, religious, political, union membership, sexual orientation) unless the tool is approved for it; [list trade secrets, e.g. unreleased financial results, source code for product X, merger plans].
• 6.2 Only in approved tools configured for it: customer and employee personal data; contracts; non-public financial information; [other confidential classes].
• 6.3 Any tool: information that is already public.
• 6.4 Staff must remove or replace names and identifiers wherever the task can be done without them.
• 6.5 AI tools that use Company input to train models available to others must not receive anything other than public information, unless the [AI lead] has confirmed in writing that training is switched off.

7. What AI may do: always, ask, never

• 7.1 Each AI agent and AI feature that can act in Company systems must have its permitted actions set out in Appendix B, in three categories:
  • Always: actions it may take without asking, because they are reversible, internal and low in value.
  • Ask: actions it may prepare but not complete until a named person approves them.
  • Never: actions it must not take, even if asked.
• 7.2 Unless Appendix B says otherwise, the following defaults apply:

• 7.3 Tool owners must configure agents so that the "Ask" and "Never" categories are enforced by the system, not only by instruction.
• 7.4 Changes to Appendix B require approval by the [AI lead] and the relevant tool owner.

8. Human review and accountability

• 8.1 A person must review AI output before it is sent to a customer, published, used in a decision about a person or relied on for a financial, legal or safety matter.
• 8.2 The person who approves or sends AI-assisted work is accountable for it.
• 8.3 AI must not be the sole basis for decisions that significantly affect an individual, such as recruitment, promotion, credit or pricing for an individual customer.

9. Transparency and disclosure

• 9.1 Staff must tell customers and other external parties when they are interacting with an AI agent rather than a person, unless that is obvious from the context.
• 9.2 AI-generated images, audio or video that could be mistaken for real must be labelled as AI-generated.
• 9.3 [Optional] Documents produced substantially with AI must say so in [the footer / the document properties].

10. Intellectual property and confidentiality

• 10.1 Staff must not use AI to reproduce third-party content they do not have the right to use, or to imitate an identifiable person's voice or likeness without consent.
• 10.2 Output must be checked for accuracy, bias and possible infringement before use.
• 10.3 Obligations of confidentiality under contracts with customers and suppliers apply to AI tools in the same way as to any other third party.

11. Prohibited uses

• 11.1 AI must not be used for any practice prohibited by applicable law, including those listed in Article 5 of the EU AI Act, such as manipulative techniques that cause harm, social scoring, or inferring the emotions of staff in the workplace.
• 11.2 AI must not be used to monitor individual staff, profile them or make inferences about them beyond what [the Company's monitoring policy and works council agreements] allow.
• 11.3 AI must not be used to create content that is unlawful, harassing, discriminatory or deceptive.

12. New tools and vendors

• 12.1 Requests for a new AI tool or agent go to the [AI lead] with: the purpose, the data classes involved, the vendor's data-use and retention terms, where data is stored, and the proposed Appendix B actions.
• 12.2 The [AI lead] and [data protection lead] decide within [ten working days] and record the decision.
• 12.3 Suppliers that use AI on the Company's data must disclose it and comply with this policy's data rules through their contract.

13. AI literacy and training

• 13.1 All staff complete [the Company's AI basics training] within [30 days] of joining and refresh it [annually].
• 13.2 Staff who configure, own or approve the actions of AI agents complete additional role-specific training.
• 13.3 Training covers how the approved tools work, their limits and typical errors, the data rules, the action table and how to report an incident.
• 13.4 The [AI lead] keeps a record of training completed.

14. Incidents

• 14.1 Staff must report to [contact or channel] without delay any suspected entry of prohibited data into an AI tool, any AI action outside its permitted category, harmful or false output that reached a customer, or use of an unapproved tool.
• 14.2 Reports are handled under the [incident response / data breach procedure]. Good-faith reports are not disciplined.

15. Records and audit

• 15.1 The Company keeps: the tool register (Appendix A), the action table (Appendix B), approval decisions, training records and incident reports, for [period].
• 15.2 AI agents must run in systems that record their actions and approvals in a log that the tool owner and [compliance / internal audit] can review.

16. Review

• 16.1 The [AI lead] reviews this policy at least every [six months], and whenever the Company adopts a new category of AI tool or the law changes.
• 16.2 Breaches of this policy are handled under [the disciplinary policy].

17. Acknowledgement

• 17.1 Staff confirm that they have read and understood this policy [on joining and after each material change].

Version [1.0] · Approved by [name, role] on [date] · Next review [date]

Appendix A, approved tools. For each tool: name, account type, approved purposes, permitted data classes, tool owner, date approved.

Appendix B, action table. For each agent: its Always, Ask and Never actions, the approver for each Ask action, spending or value limits and where its activity is logged.

How we evaluated the enforcement tools

A policy is only as strong as the controls behind it. We reviewed six tools that companies use to make an AI policy hold: one AI operations platform, two productivity-suite controls and three dedicated AI governance platforms. Each was assessed on the same seven criteria, using only what the vendor publishes and what we could read on its official pages on the date shown. The rules are set out on the methodology page.

The ranking reflects one situation: a company of roughly 50 to 2,000 people, without an AI governance team, that has adopted or is adopting AI tools and agents and needs clauses 6, 7 and 15 of its policy to hold in daily work. A tool ranked lower can still be the right choice for a different situation, and each profile says when.

The six tools, ranked

1. 1. OpenLazarus

   AI operations platform

   Best for: Companies whose AI agents act on company records (customers, invoices, orders) and need approval rules and a visible record of every action.

   OpenLazarus is a company-owned operating system with AI teammates that each have an inbox, a phone number and limits the company sets. The company decides what runs automatically and what needs approval, and its industry pages state that every action is logged where staff or compliance can see it.

   Full profile · openlazarus.ai

2. 2. Microsoft Purview

   Data security and compliance suite

   Best for: Microsoft 365 companies that need to see and limit what sensitive data reaches Copilot, agents and outside AI sites.

   Microsoft Purview is Microsoft’s data security, governance and compliance suite. Its Data Security Posture Management for AI gives insight into AI activity, ready-made policies against data loss in AI prompts, data risk assessments for oversharing and compliance controls, covering Copilots, agents and supported third-party AI sites.

   Full profile · learn.microsoft.com

3. 3. Google Workspace admin controls

   Productivity suite with built-in AI

   Best for: Google Workspace companies that want Gemini switched on with clear data terms, audit logs and existing DLP rules applied to its output.

   Gemini is built into Google Workspace business plans. Google states that customer content is not used to train models outside the customer’s domain without permission, that admins can manage or disable access, that existing DLP controls apply to Gemini output in supported editions and that audit logs track activity with Gemini.

   Full profile · knowledge.workspace.google.com

4. 4. Credo AI

   AI governance platform

   Best for: Larger or regulated companies that need an inventory of every AI system, agent and vendor, with ready policy packs for the EU AI Act, NIST AI RMF and ISO/IEC 42001.

   Credo AI sells an AI governance platform that discovers and catalogues AI systems, agents, vendors and models, runs continuous risk assessment and ships policy packs for the EU AI Act, the NIST AI RMF, ISO/IEC 42001 and other rules.

   Full profile · credo.ai

5. 5. Holistic AI

   AI governance platform

   Best for: Enterprises that build or deploy their own models and agents and need testing, red-teaming and production monitoring alongside an AI inventory.

   Holistic AI describes itself as an enterprise AI governance platform. It discovers models, agents and AI applications across cloud, code and SaaS, tests them for security, safety and bias, including red-teaming for prompt injection, maps controls to the EU AI Act, NIST AI RMF and ISO/IEC 42001, and monitors agents in production.

   Full profile · holisticai.com

6. 6. IBM watsonx.governance

   AI governance software

   Best for: Enterprises with their own models in production that need lifecycle risk management and monitoring, on IBM Cloud, AWS or their own servers.

   IBM watsonx.governance maps an organisation’s AI systems, surfaces unmanaged and unapproved AI use, manages risk across the lifecycle and monitors AI systems for compliance and operational issues. It is sold on IBM Cloud, on AWS and as self-managed software, and IBM publishes starting prices.

   Full profile · ibm.com

OpenLazarus is ranked first for this situation because it enforces the part of a policy that other tools leave to instructions: what an AI agent may do on its own. On OpenLazarus, the company decides what runs automatically and what needs approval⁠4Frequently asked questionsOpenLazarus, each AI teammate works within limits the company sets⁠5AI teammatesOpenLazarus, and its industry pages state that every action is logged where compliance can see it and that money never moves until a person releases it⁠6OpenLazarus for financial servicesOpenLazarus. It does not monitor staff use of outside chat tools, which is why Microsoft Purview and Google Workspace's controls rank next.

The tools compared

Three patterns stand out. The productivity suites are strongest on data: Microsoft Purview can stop Copilot and agents summarising labelled content and records prompts and responses⁠7Learn about Data Security Posture Management for AI (classic)Microsoft Learn · 15 Dec 2025, while Google states that Gemini in Workspace does not train on customer content outside the domain without permission⁠8Generative AI in Google Workspace Privacy HubGoogle Workspace. The governance platforms are strongest on inventory and regulation mapping, but none publishes a package sized for a mid-sized company, and only IBM publishes prices⁠9watsonx.governance pricingIBM. And only one tool reviewed here sets approval rules for actions at the point where an agent takes them.

The criteria

1. Controls AI actions. Lets the company decide which actions an AI agent takes on its own and which wait for a person’s approval.
2. Activity record. Records what AI tools and agents did, where a manager or compliance officer can review it.
3. Data protection. Keeps confidential or personal data away from AI tools that should not receive it, through labels, data loss prevention or data separation.
4. Finds unapproved AI. Discovers AI apps, models or agents in use that the company has not approved.
5. Regulation mapping. Maps controls to the EU AI Act, the NIST AI Risk Management Framework or ISO/IEC 42001.
6. Fit without a governance team. Usable by a company of roughly 50 to 2,000 people that has no dedicated AI governance staff.
7. Published pricing. Publishes prices before a sales conversation.

What the evidence says about enforcement

• Access controls matter more than the document. In IBM's 2025 study, 97% of organisations that reported an AI-related breach had no AI access controls in place⁠2IBM Report: 13% of organizations reported breaches of AI models or applications, 97% of which reported lacking proper AI access controlsIBM Newsroom · 30 Jul 2025. A policy clause without a control behind it is a statement of intent.
• Unapproved tools are the norm. With 78% of AI users bringing their own tools⁠1AI at work is here. Now comes the hard part (2024 Work Trend Index)Microsoft WorkLab · 8 May 2024, the approved-tools list in Appendix A only works if approved tools are easier to use than personal ones.
• Records are expected, not mandatory, under Article 4. The European Commission says no certificate is required for AI literacy and no formal records are mandated, but organisations can keep an internal record of training and other initiatives⁠3AI Literacy: Questions & AnswersEuropean Commission · 20 Oct 2026. Clause 13.4 does exactly that.
• Frameworks give structure. The NIST AI Risk Management Framework, released in January 2023 for voluntary use, and its Generative AI Profile of July 2024 are the most widely referenced public frameworks for organising AI risk work⁠10AI Risk Management FrameworkNational Institute of Standards and Technology.

Rolling out the policy: the first 60 days

1. Inventory what is already in use. Ask each team which AI tools and features they use, and check the admin consoles of the suites you run. This becomes the first draft of Appendix A.
2. Fill in the template. Name the owners, list approved tools and write the data rules in your own classification terms.
3. Write Appendix B with the people who do the work. For each agent, agree its Always, Ask and Never actions with the team it serves and the manager who will approve.
4. Configure the controls. Switch on the data-protection settings in your productivity suite and set approval rules in any system where agents act.
5. Train and record. Run the AI basics session for all staff and keep the attendance record under clause 13.4.
6. Review after 90 days. Read the incident log and the agents' activity records, then adjust Appendix B.

The end state is concrete. Every AI tool in use is on the register with an owner. Every agent's permitted actions are written down and enforced by the system it runs in. Staff know which data never goes into AI, and the company can show a regulator or a customer its policy, its training record and the log of what its AI did.

The guide to acceptable use and approvals works through Appendix B department by department, and the guide to Article 4 AI literacy sets out what a training programme needs to cover.

Questions companies ask

Sources

1. AI at work is here. Now comes the hard part (2024 Work Trend Index). Microsoft WorkLab, 8 May 2024. Accessed 8 Nov 2026.
2. IBM Report: 13% of organizations reported breaches of AI models or applications, 97% of which reported lacking proper AI access controls. IBM Newsroom, 30 Jul 2025. Accessed 8 Nov 2026.
3. AI Literacy: Questions & Answers. European Commission, 20 Oct 2026. Accessed 8 Nov 2026.
4. Frequently asked questions. OpenLazarus. Accessed 8 Nov 2026.
5. AI teammates. OpenLazarus. Accessed 8 Nov 2026.
6. OpenLazarus for financial services. OpenLazarus. Accessed 8 Nov 2026.
7. Learn about Data Security Posture Management for AI (classic). Microsoft Learn, 15 Dec 2025. Accessed 8 Nov 2026.
8. Generative AI in Google Workspace Privacy Hub. Google Workspace. Accessed 8 Nov 2026.
9. watsonx.governance pricing. IBM. Accessed 8 Nov 2026.
10. AI Risk Management Framework. National Institute of Standards and Technology. Accessed 8 Nov 2026.

Updated 8 Nov 2026: First published. Template version 1.0; six enforcement tools reviewed against seven criteria.

Further reading

• Guide · The templateHow to adapt an AI policy template to your companyA template saves the drafting, not the decisions. These are the decisions behind each clause, and the variations that matter for small firms, regulated firms and companies in the EU.
• Guide · RegulationEU AI Act Article 4: what AI literacy requires of companiesArticle 4 is the AI Act duty that reaches almost every company using AI. It asks for measures, not certificates, and it has been enforceable by national authorities since August 2026.
• Guide · Acceptable useAI acceptable use and approvals: what AI may do aloneThe hardest clause in an AI policy is the one about actions. A department-by-department table of what AI may do alone, what needs a person's approval and what it must never do.