Generative AI Acceptable Use Policy: Sections Every Employer Needs

Photo: Pixel.la Free Stock Photos, CC0 (public domain), via Wikimedia Commons.
At most companies of any size, someone has probably already pasted a customer email, a draft performance review or a chunk of source code into a public chatbot. Not out of malice. It saved twenty minutes. A generative AI acceptable use policy exists to make that moment safe: it tells people which tools are approved, what data stays out, and who checks the output before it matters.
A generative AI acceptable use policy tells employees which AI tools they may use, what data they may and may not enter, how they must check outputs, and what to do when something goes wrong. Build it around the risks NIST names in its Generative AI Profile (NIST AI 600-1), such as confabulation, data privacy, information security and human-AI configuration, and keep it short enough that people actually read it.
By the Decision Makers Hub editorial team. Last updated November 2026. This article is general information, not legal advice. Have counsel review any policy before you adopt it, especially for hiring, privacy and regulated data.
Key takeaways
- Use NIST's twelve generative AI risks as a checklist for your policy sections. NIST AI 600-1 was approved for publication in July 2024.
- NIST's suggested actions include acceptable use policies (GV-1.4-002) and after-action reviews of AI incidents (GV-1.5-002).
- The data section does the most work. Spell out what may never be entered, and name the approved tools.
- Treat AI output as a draft. NIST defines confabulation as "confidently stated but erroneous or false content."
- Put people decisions (hiring, firing, discipline, pay) behind a human reviewer and your legal checklist.
What we keep hearing. A vendor says its AI features are "enterprise grade" and "compliant," and the policy question gets treated as settled.
Our take. Ask in writing whether your prompts are stored or used to train models, and get the answer into the contract. A sales deck is not a policy.
Why write a separate policy for generative AI?
A general "AI policy" often covers strategy, vendors and governance. An acceptable use policy is narrower and more practical: it is the page an employee reads before opening a chatbot. If you already have a company-wide document, such as the one in our company-wide AI policy template, this policy is the employee-facing layer that sits under it. For the program above both, see how leaders set up an AI governance framework.
The NIST AI Risk Management Framework, released January 26, 2023, is voluntary, and its Generative AI Profile (NIST AI 600-1, July 2024) applies it to tools that generate text, images and code. NIST describes the profile's suggested actions in terms such as: "Establish transparent policies that address illegal uses or applications of GAI" (GV-1.4-002).
Which risks should the policy cover?
NIST lists twelve generative AI risks. Four of them map directly onto what employees do all day.
| NIST risk | NIST description (short) | Policy section it belongs in |
|---|---|---|
| Confabulation | "The production of confidently stated but erroneous or false content" | Accuracy and verification |
| Data Privacy | Leakage and unauthorized use or disclosure of personal or sensitive data | Data rules |
| Information Security | Lowered barriers for offensive cyber capabilities and a larger attack surface | Security and access |
| Human-AI Configuration | Interactions between a human and an AI, including over-reliance and automation bias | Human review and high-stakes decisions |
The other eight (CBRN information, dangerous or hateful content, environmental impacts, harmful bias, information integrity, intellectual property, obscene or abusive content, and value chain integration) can be handled with a short "prohibited uses" list and a link to your IP and ethics policies. For how bias shows up in hiring tools specifically, read our guide to the NYC Local Law 144 bias audit.
What sections should the policy include?
- Purpose and scope. State who it applies to (employees, contractors, interns) and what counts as generative AI (chatbots, writing assistants, code assistants, image generators, AI features inside existing software).
- Approved tools and access. List the tools employees may use and how they sign in. Say plainly that company data does not go into personal accounts or unapproved tools.
- Data rules. Name what may never be entered: customer or employee personal data, health or financial information, credentials, unreleased financials, and confidential source code unless a tool is approved for it. Our guide to protecting employee data in HR systems lists the data classes worth naming.
- Accuracy and verification. Require a person to check facts, figures, quotes and citations before anything leaves the building. Never paste AI output into legal, financial or policy documents unchecked.
- High-stakes people decisions. Prohibit using generative AI as the sole basis for hiring, promotion, discipline, termination or pay. Require review by a trained human, as we discuss in human in the loop AI.
- Intellectual property and disclosure. Say who owns outputs, when to disclose AI assistance to customers or candidates, and what to do about third-party copyrighted material.
- Security. Require approved integrations only, and tell people to report suspicious tool behavior. Treat prompts and outputs as business records.
- Incident reporting. Give a named contact and a deadline for reporting a leak, a harmful output or a policy breach. NIST suggests after-action reviews of AI incident response (GV-1.5-002), so plan to learn from each report.
- Training and enforcement. Tie acceptance of the policy to onboarding and annual training, and state the consequences of violations.
- Review cadence. Set a date to revisit the policy. Tools and rules change faster than most handbooks.
What does sample language look like?
Adapt these lines with counsel. They are starting points, not legal text.
Data: "Do not enter personal data about employees, candidates or customers, credentials, or confidential business information into any AI tool that the company has not approved in writing for that purpose."
Accuracy: "AI output can be wrong while sounding certain. You are responsible for checking facts, numbers and sources before you use or share anything an AI tool produces."
People decisions: "AI tools may not make or be the sole basis for decisions about hiring, promotion, pay, discipline or termination. A trained human reviewer must make and document these decisions."
Reporting: "Report a suspected data leak or harmful AI output to the designated contact within one business day."
How do you roll it out so people follow it?
- Find the AI you already have. Survey teams and check your software list. Our guide to auditing your HR tech stack shows how to find tools nobody formally approved.
- Approve a small set of tools first. A banned list with no approved alternative just moves usage out of sight.
- Train with real examples. Show a safe prompt and an unsafe prompt for each team.
- Make reporting easy and blameless. People report mistakes faster when the response is learning, not punishment.
- Review on a schedule. Add a calendar date and an owner. Distinguishing between machine learning, deep learning and generative tools also helps non-technical staff, so link them to AI vs machine learning vs deep learning.
Frequently asked questions
What should a generative AI acceptable use policy include?
Scope, approved tools, data rules, accuracy checks, limits on high-stakes people decisions, IP and disclosure, security, incident reporting, training and enforcement, and a review date.
Is the NIST Generative AI Profile mandatory?
No. NIST says the AI Risk Management Framework is intended for voluntary use. The Generative AI Profile (NIST AI 600-1) is a companion resource that applies the framework to generative AI.
Should employees use personal AI accounts for work?
Most policies say no. Company data should go only into approved tools accessed through company accounts, so the organization can control retention and access.
How often should the policy be reviewed?
Set a fixed cadence, such as every six months, and review sooner after an incident or a major tool change. NIST's suggested actions include after-action reviews of AI incident response.
Does the policy need legal review?
Yes. Privacy, employment and sector-specific rules vary, so have counsel review the final text.
Comments
Post a Comment