Free template
AI security policy template for organisations
A practical AI acceptable use and security policy you can adapt today. It covers approved tools, what data may go into AI, AI coding assistants, agents and MCP integrations, prompt injection and incident reporting.
Why you need one
Your teams are already using AI. A policy makes it safe to say yes.
Without clear rules, employees paste source code and customer data into public AI tools, install unvetted AI extensions and connect agents to production systems. An AI security policy sets the boundaries, and gives security teams something to enforce.
This template is written by the PexLens research team and reflects real incidents in AI tooling, including malicious updates to AI coding extensions. Replace the bracketed text with your own details.
The template
AI acceptable use and security policy
1. Purpose
This policy sets out how [Company] employees, contractors and systems may use artificial intelligence (AI) tools, including generative AI assistants, AI coding assistants, AI browsers and AI agents, while protecting company data, customers and systems.
2. Scope
This policy applies to all AI tools used for company work on any device, including free and paid services, browser extensions, IDE extensions, plugins, MCP servers and AI features built into other software.
3. Roles and ownership
- Security: owns this policy, approves AI tools and monitors usage.
- Engineering leads: approve AI coding tools and agent integrations for their teams.
- Legal and privacy: review data protection and contractual terms of AI vendors.
- All staff: follow this policy and report incidents.
4. Approved AI tools
Only AI tools on the approved list may be used with company data. Requests for new tools go to Security, which reviews the vendor's data handling, retention, training use and access model. Personal AI accounts must not be used for company work.
5. Data rules
| Data type | Allowed in approved AI tools? |
|---|---|
| Public information | Yes |
| Internal, non-sensitive information | Yes, in approved tools only |
| Source code | Only in approved coding tools configured not to retain or train on data |
| Customer data and personal data (PII) | No, unless explicitly approved for the use case |
| Secrets, credentials, API keys | Never |
| Regulated data (financial, health) | Never, unless approved by Legal and Security |
6. AI coding assistants and extensions
- Install AI coding assistants, IDE extensions and plugins only from approved publishers.
- Updates to approved AI tools are subject to review; tools that add new permissions, network destinations or behaviour may be quarantined.
- AI-generated code is reviewed by a human before merge, like any other code.
- AI tools must not have write access to protected branches, CI/CD configuration or production secrets without approval.
7. AI agents and integrations
- Each agent or MCP integration has a named owner and a documented list of what it can read, change and call.
- Agents use scoped, short-lived credentials, never a person's full access.
- High-risk actions, such as deploying, deleting data or changing access, require human approval.
8. Security controls
The company may inspect, log and block AI traffic from company devices to enforce this policy, including redacting sensitive data from prompts. Logs are kept for [retention period] and handled under the company privacy policy.
9. Prompt injection and output risk
Treat AI output as untrusted. Do not run commands, open links or approve actions suggested by an AI tool without checking them, especially when the tool has read external content such as web pages, emails or third-party code.
10. Incidents
Report suspected data exposure through AI tools, malicious AI extensions or unexpected agent behaviour to Security immediately at [contact].
11. Training
All staff complete AI security awareness training when they join and every year after.
12. Exceptions and review
Exceptions require written approval from Security. This policy is reviewed every [6 or 12] months and whenever a significant new AI capability is adopted.
From policy to practice
A policy only works if it is enforced
PexLens turns this policy into controls: AI Guard redacts or blocks sensitive data in prompts, and PexLens vets AI extensions, packages and agent components before they reach developers.
FAQ
AI security policy: common questions
What should an AI security policy include?
At minimum: which AI tools are approved, what data may and may not be shared with them, rules for AI coding assistants, extensions and agents, how usage is monitored, how to report incidents, training requirements and how often the policy is reviewed.
What is the difference between an AI acceptable use policy and an AI security policy?
An acceptable use policy tells employees how they may use AI. An AI security policy adds the technical controls, ownership and incident handling that protect data and systems. This template combines both.
Should we ban ChatGPT and other AI tools?
Outright bans tend to push usage to personal accounts and devices. Most organisations get better results by approving specific tools, setting data rules and enforcing them with inspection and logging.
Who should own the AI security policy?
Usually the security team or CISO, working with legal, privacy, IT and engineering. Review it at least once a year and whenever you adopt a significant new AI tool.
How do we enforce an AI security policy?
Combine an approved-tools list and training with technical controls: AI Guard to inspect, redact or block sensitive data in prompts, PexLens policies for AI-related extensions and packages, and logging so you can audit usage.
Can I use this template for free?
Yes. Copy the AI security policy template, adapt it to your organisation and share it with your teams.
Related
More on AI security
Enforce your AI policy automatically
See how PexLens applies AI usage policy to prompts, extensions and agents.