Back to Insights
AI Governance8 min read11 November 2026

Generative AI Acceptable-Use Policy: What to Include

A generative AI acceptable-use policy is the control that answers shadow AI: it tells staff which tools they may use, what data may go in, how outputs must be checked and what is off limits. Here is a section-by-section checklist of what to include, anchored in NIST's generative AI profile and OWASP's guidance on sensitive information disclosure.

AA

Agraj Agranayak

Founder & CEO, Imagine Works · About · LinkedIn

Key Takeaways

  • An acceptable-use policy is the day-to-day control for generative AI. NIST's Generative AI Profile suggests establishing transparent acceptable use policies, and updating them to cover third-party personnel and both proprietary and open-source tools.
  • The data rule does the most work. OWASP lists personal data, financial details, health records, confidential business data, security credentials and legal documents as sensitive information that generative AI applications can expose, so a policy needs a plain list of what must never be entered.
  • Outputs need a human check. NIST describes confabulation as generative AI confidently presenting erroneous or false content, so the policy should say who reviews AI output, for which uses, before it is relied on or sent.
  • A policy without an approved route gets bypassed. Pair the policy with a list of approved tools and a simple way to request a new one, and link it to your AI system inventory.
  • Have it reviewed, and revisit it. Counsel and HR should review the final wording, and the policy should have a named owner and a review date, because tools and rules change.

Most organisations have staff using generative AI tools today, whether or not anyone has told them what is acceptable. A generative AI acceptable-use policy is the document that does. It is short, it is read by every employee, and it is the control that sits closest to how the technology is actually used.

This piece sets out what such a policy should cover, section by section. It is advisory: it describes the content a policy should have and the decisions behind each section. Your counsel and HR function should review the final wording for your jurisdiction and your workforce.

Why a Policy, and Why Now

The 2026 conversation about unsanctioned AI use is covered in our guide to shadow AI governance. The short version: if staff use tools you have not sanctioned, you cannot protect the data that goes into them. A policy is the first response, and it works only if it is specific.

NIST's Generative AI Profile (NIST AI 600-1, July 2024) points the same way. Among its suggested actions it asks organisations to establish transparent acceptable use policies for generative AI that address illegal use or applications; to define acceptable use policies for generative AI interfaces and human-AI configurations, including criteria for the kinds of queries a generative AI application should refuse; and to update acceptable use policies to address proprietary and open-source generative AI technologies and data, and contractors, consultants and other third-party personnel (GV-1.4-002, GV-3.2-003 and GV-6.1-010, accessed 16 November 2026). NIST adds that acceptable use policies and guidance for generative AI in formal human-AI teaming settings, as well as in different levels of human-AI configurations, can help decrease risks arising from misuse, abuse, inappropriate repurposing and misalignment between systems and users.

What the Policy Should Contain

1. Purpose, scope and who it applies to

Say in two sentences why the policy exists, and who it covers: employees, contractors and consultants. NIST's suggested action about third-party personnel is a reminder that the people using AI on your behalf are not all on your payroll. State which tools are in scope: tools the company provides, tools built into software you already use, and tools staff use on their own accounts for work.

2. Approved tools, and how to get a new one approved

List the tools staff may use, or point to where the list lives. This should link to your AI system inventory: the inventory is the record of what is approved and for what use. Then say how a person requests a new tool, who decides, and roughly how long it takes. Our advice is to pair every restriction with an approved alternative, because a policy that only says "no" is more likely to be bypassed.

3. Data rules: what must never go in

This is the section that does the most work, so make it concrete. OWASP's 2025 Top 10 for LLM Applications describes sensitive information as including personally identifiable information, financial details, health records, confidential business data, security credentials and legal documents, and warns that generative AI applications can expose sensitive data through their output (OWASP, LLM02:2025 Sensitive Information Disclosure, accessed 16 November 2026). NIST's profile similarly notes that models may leak, generate or correctly infer sensitive information about individuals.

Turn that into a plain list for your organisation, for example:

  • Personal data about customers, employees or candidates.
  • Client or partner confidential information.
  • Credentials, keys, tokens and passwords.
  • Source code or documents under confidentiality or legal privilege, unless the specific tool is approved for that class of data.
  • Anything your own data classification scheme labels confidential or above.

Also say what is allowed. Staff are more likely to follow a rule that tells them what they can do, such as drafting from public information, summarising non-confidential text or brainstorming.

One caution that belongs in the policy's rationale. OWASP notes that restricting what a model returns through system prompt instructions can reduce disclosure risk but may not always be honoured and can be bypassed. A policy that relies on staff and tool settings alone is weaker than one that also limits which tools may see which data.

4. Output rules: check before you rely or send

NIST describes confabulation as generative AI systems generating and confidently presenting erroneous or false content in response to prompts. The policy should say:

  • Who is accountable for AI-assisted work. The person who uses AI output remains responsible for it.
  • What must be reviewed, and by whom, before it is relied on, published or sent outside the organisation. A reasonable tiering: low-stakes internal drafts need the author's review; anything customer-facing, financial, legal, or affecting a person needs a second, qualified reviewer.
  • When AI use must be disclosed, to colleagues, clients or the public, if your sector or contracts require it.
  • Decisions about individuals. Say whether AI output may inform decisions about people, and require a human who can override. Our guide to human-in-the-loop design covers what meaningful oversight looks like.

5. Intellectual property and licensed content

NIST's profile notes that how generative AI relates to copyright, including the status of generated content similar to protected work, is currently being debated in legal fora, and that similar debates cover the use of personal identity, likeness or voice without permission. Because the position is unsettled, the policy should be cautious: do not enter third-party copyrighted or licensed material into a tool unless its licence and the tool's terms allow it, and have legal review any use of AI-generated content in published or client-facing work.

6. Prohibited uses

State what is off limits regardless of tool. NIST's first suggestion is to address illegal use or applications. Beyond that, list the uses your organisation will not allow, for example: generating deceptive content about real people, bypassing security controls, using unapproved tools for regulated or confidential work, or connecting an AI tool to company systems without approval. If your organisation builds or deploys its own chatbots, NIST's suggestion to define criteria for the kinds of queries an application should refuse belongs in a companion standard for the teams that run them.

7. Security behaviour

NIST's profile notes that generative AI expands the attack surface, being vulnerable to attacks such as prompt injection and data poisoning. For staff, this translates into simple rules: treat content pulled into an AI tool from outside (emails, web pages, documents) as potentially hostile, do not give a tool access to company systems or accounts without approval, and report anything odd. Our guide to prompt injection explains the risk for leaders.

8. Training

A policy nobody has read is not a control. Link the policy to training, and keep the training matched to the role. We cover the wider topic in our guide to AI literacy.

9. Reporting and incidents

Tell staff how to report a mistake, a leak, or a suspected misuse, and make it safe to do so. Point to your AI incident response process so the report goes somewhere real.

10. Ownership, exceptions, consequences and review

Name an owner. Say how exceptions are requested and recorded. State that breaches are handled under your existing disciplinary process, and have HR and counsel confirm the wording. Set a review date, because tools and rules change; a policy written for chatbots will not cover agents that take actions.

The Checklist on One Page

  • Scope — Who and which tools does this cover?
  • Approved tools — Which tools may I use, and how do I ask for another?
  • Data rules — What must I never enter? What can I enter?
  • Output rules — Who checks AI output, and when?
  • IP — Can I put third-party material in? Can I publish what it produces?
  • Prohibited uses — What is off limits whatever the tool?
  • Security — What should I treat as hostile, and what must I not connect?
  • Training — Where do I learn this?
  • Reporting — Who do I tell, and how?
  • Ownership and review — Who owns this, and when is it next reviewed?

What Leaders Should Be Asking

  • Do we have a written generative AI policy, and has every employee and contractor seen it?
  • Does it contain a concrete list of data that must never be entered, in language staff can apply?
  • Is there an approved list of tools and a quick way to request a new one?
  • Who reviews AI output before it reaches a customer, and is that written down?
  • When was the policy last reviewed, and who owns the next review?

The Underlying Point

A generative AI acceptable-use policy cannot prevent every mistake. Its job is to make the right behaviour easy to understand, give staff an approved path, and give the organisation something to enforce and improve. Short, specific and reviewed beats long and generic.

Imagine Works advises enterprise leaders on AI governance and risk design, including how to scope an acceptable-use policy, connect it to your AI inventory and risk process, and decide who owns it. We advise; we do not build or implement software. Get in touch to discuss your AI governance programme.

Sources (as of 16 November 2026)

All opened and read on 16 November 2026.

  1. 1NIST AI 600-1, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, July 2024 (suggested actions GV-1.4-002, GV-3.2-003, GV-6.1-010; risk descriptions for confabulation, data privacy, information security, intellectual property; the closing remark on acceptable use policies and human-AI configurations)
  2. 2OWASP Gen AI Security Project, LLM02:2025 Sensitive Information Disclosure
  3. 3OWASP Gen AI Security Project, 2025 Top 10 Risk & Mitigations for LLMs and Gen AI Apps (index page)

Related Service

AI Governance & Risk Design

Designing the governance framework and risk architecture that keeps your AI systems compliant, auditable, and board-ready — before regulation forces the issue.

Explore this service

More Insights

More on AI Governance

View all
AI Governance8 min read

How to Build an AI Risk Register: What to Record and Who Owns It

An AI risk register is the working record of what could go wrong with each AI system, how serious it is, who owns it and what is being done. NIST's AI Risk Management Framework does not use the term, but it asks organisations to document exactly the things a register holds. Here are the fields, the owners, the review rhythm and the common failures.

16 November 2026Read article
AI Governance10 min read

The NIST AI Risk Management Framework: What Enterprise Leaders Need to Understand

The NIST AI Risk Management Framework (AI RMF 1.0) has quietly become the most widely adopted enterprise AI governance framework in the United States — 57–67% of CISOs use it according to the 2026 Hitch Partners Global CISO Leadership Report. Voluntary, but load-bearing. Here is what the framework contains, how the 2024 Generative AI Profile extends it, and why it endured a change of US administration when the executive order that popularised it did not.

24 June 2026Read article
AI Governance10 min read

Prompt Injection: The Enterprise Security Risk That Cannot Yet Be Filtered Away

Prompt injection has been the number-one risk on the OWASP Top 10 for LLM Applications since 2023, and in June 2025 it produced the first publicly documented zero-click enterprise AI vulnerability — CVE-2025-32711, EchoLeak, in Microsoft 365 Copilot. UK NCSC's December 2025 guidance is direct: prompt injection cannot be fully mitigated; focus on reducing impact. Here is what enterprise leaders need to understand and do.

10 June 2026Read article