However, a policy alone is just a document. The organizations that actually reduce risk pair a clear, short policy with access controls that enforce it automatically. This guide covers both: what to put in your AI acceptable use policy, a template you can adapt, and how zero trust principles turn the policy into something your network actually follows.
What is an AI acceptable use policy?
An AI acceptable use policy (AI AUP) is a written set of rules that defines how employees, contractors, and automated systems may use artificial intelligence tools for business purposes. It typically covers which tools are approved, what data can and can't be entered into them, who is accountable for AI-generated output, and how violations are handled.
Most organizations write one in response to shadow AI: the use of AI tools and services that IT or security hasn't approved. When there's no policy, people make their own decisions about which chatbot, coding assistant, or browser extension to trust with company data. The AI AUP replaces those individual guesses with one shared standard.
Why every company needs one now
- Data leaves your control the moment it's pasted into a prompt. Customer records, source code, contracts, and financials can end up with a third-party model provider whose retention and training terms you never reviewed.
- Compliance obligations don't pause for new tools. If you're subject to GDPR, HIPAA, SOC 2, PCI DSS, or customer contract terms, you need to show where regulated data goes. Unsanctioned AI creates gaps auditors will ask about.
- AI agents act on their own. Agents and automations can call external AI services from servers and pipelines without a human in the loop, so the policy has to cover machines, not just people.
- Employees want guidance. Most people aren't trying to break rules. They use whatever helps them get work done. A clear policy with an approved alternative removes the guesswork.
What to include in an AI acceptable use policy
Keep it short enough that people will actually read it. These are the sections that matter most.
1. Purpose and scope
State why the policy exists and who it applies to: employees, contractors, vendors with system access, and any AI agents or automations running on company infrastructure.
2. Approved, restricted, and prohibited tools
List the AI tools that are approved (ideally enterprise tiers with data protection terms), tools that are allowed only for certain teams or use cases, and tools that are blocked. Name a single place where the current list lives, because it will change often.
3. Data classification rules
Map your existing data classes to AI use. For example: public information may be used with any approved tool; internal information only with approved enterprise tools; confidential, customer, or regulated data never, unless a specific tool has been approved for it.
4. Human review and accountability
Make clear that the person using the tool owns the output. AI-generated code, customer communications, and legal or financial content need human review before they're used.
5. Requesting a new tool
Give people a fast, simple way to request a tool that isn't on the list. If the approval path is slow, people work around it, and that's how shadow AI grows.
6. AI agents and automations
Require that every agent has a named owner, a documented purpose, and access only to the systems and AI services it needs. Our guide to giving AI agents least-privilege access goes deeper on this.
7. Monitoring and enforcement
Tell people that AI usage is monitored at the network level and that unapproved services are blocked. Transparency here builds trust and reduces accidental violations.
8. Violations and review cadence
Explain what happens when the policy is broken, and commit to reviewing the policy and the approved-tools list regularly, such as quarterly.
AI acceptable use policy template
To help you kick off your acceptable use policy, we have created a template to tailor to your org. Adapt the bracketed sections to fit your organization. This is a starting point, not legal advice; have your legal and compliance teams review your final version.
[Company] AI Acceptable Use Policy
- Purpose. This policy sets rules for the use of artificial intelligence tools to protect [Company] data, customers, and systems while allowing employees to benefit from AI.
- Scope. Applies to all employees, contractors, and third parties with access to [Company] systems, and to all AI agents and automations running on [Company] infrastructure.
- Approved tools. Only tools listed on the [Approved AI Tools page] may be used with company data. All other AI services are blocked on company networks and devices.
- Data rules. Do not enter confidential, customer, personal, or regulated data into any AI tool unless that tool is approved for that data class.
- Output. You are responsible for reviewing and verifying AI-generated content before it is used, shared, or deployed.
- New tools. Request new tools through [request process]. IT and Security will respond within [X business days].
- AI agents. Every agent must have a named owner, documented purpose, and least-privilege access to systems and AI services.
- Monitoring. [Company] monitors network access to AI services and blocks unapproved services.
- Violations. Violations may result in [consequences], consistent with [Company] policies.
- Review. This policy and the approved-tools list are reviewed every [quarter].
Why a written policy isn't enough on its own
A policy tells people what to do. It can't stop a rushed employee from pasting a customer list into an unapproved chatbot, and it can't stop an AI agent from calling a service nobody signed off on. Policies that depend entirely on trust leave you with no enforcement and no visibility.
That's where zero trust comes in. Zero trust assumes no user, device, or workload should be trusted by default. Every connection is verified, and access is limited to what's explicitly allowed. Applied to AI, that means your approved-tools list stops being a document and becomes a rule your network enforces.
How zero trust access controls enforce your AI policy
Each section of your AI AUP maps to a control. Here's how that works with CloudConnexa, OpenVPN's cloud-delivered Zero Trust Network Access platform.
Block unapproved AI services with domain filtering
Cyber Shield, built into CloudConnexa, lets you maintain custom block and allow lists alongside dozens of content categories. Add the AI domains your policy prohibits to a block list, assign it to the right user groups, and the "prohibited tools" section of your policy is enforced everywhere users connect. See how DNS security and content filtering works, or follow the step-by-step setup in How to Control Shadow AI.
Apply different rules to different teams
Policies often allow certain tools for specific roles, like a code assistant for engineering only. Group-based policies let you apply restricted-tool rules to the teams that need them without opening access for everyone.
Only let trusted devices reach company data
Your data rules mean little if a personal, unpatched laptop can connect. Device posture checks confirm that devices meet your security requirements before they can connect and while they stay connected.
Cover AI agents, not just people
Agents running on servers and in cloud networks connect through the same zero trust fabric, so the same block lists and least-privilege rules apply to them. That closes the gap between what your policy says about agents and what they can actually reach.
Prove the policy is working
Enforcement without evidence won't satisfy an auditor. The DNS log shows which domains users and agents try to reach, and log and audit streaming sends that activity to your SIEM. Blocked requests also show you which tools people want, which is useful input for your next approved-tools review.
Protect the SaaS apps you do approve
Approved AI tools still hold your data. SaaS and cloud protection lets you restrict access to sanctioned apps so only verified users on trusted devices can reach them.
Policy-to-control checklist
- Approved and prohibited tools are listed in one place, and prohibited domains are on a block list.
- Restricted tools are assigned only to the groups that need them.
- Data rules are backed by device posture checks for anything that can reach sensitive systems.
- Every AI agent has an owner, a purpose, and least-privilege network access.
- DNS and access logs stream to your SIEM, and blocked-request data feeds your quarterly policy review.
- Employees know the policy exists, how to request a tool, and that enforcement is automatic.
Frequently asked questions
What should an AI acceptable use policy include?
At minimum: scope, approved and prohibited tools, data classification rules, accountability for AI output, a process for requesting new tools, rules for AI agents, monitoring and enforcement, and a review schedule.
Is an AI acceptable use policy the same as an AI governance policy?
Not quite. AI governance is the broader program covering risk assessment, model selection, ethics, and oversight. The acceptable use policy is the employee-facing part of that program that sets day-to-day rules.
How do you enforce an AI acceptable use policy?
Combine training and clear communication with technical controls: domain filtering to block unapproved AI services, group-based access policies, device posture checks, and logging that shows what's being accessed. Zero trust access platforms like CloudConnexa bring these controls together.
Does an AI acceptable use policy need to cover AI agents?
Yes. Agents and automations can send data to external AI services without a person involved, so they need owners, documented purposes, and least-privilege access just like users do.
How often should you update your AI acceptable use policy?
Review it at least quarterly. New AI tools appear constantly, and your approved-tools list should keep pace.