To write an AI use policy, decide who owns AI decisions, list the AI tools and uses already in your business, sort them by risk, then set clear rules on approved tools, data that must never be shared, human review and disclosure. Wrap those rules in a light governance framework, such as one based on the NIST AI Risk Management Framework, so the policy is enforced, measured and updated.
Many businesses already have AI in daily use without any written rules. This guide explains the difference between an AI policy and an AI governance framework, how to use the NIST framework without turning it into a bureaucracy, what a practical policy should contain and which laws and standards are worth knowing about. It is written for owners and executives of small and mid-sized US businesses, not for compliance specialists.
Why your business needs an AI policy now
The main reason is simple: your people are probably already using AI. In Microsoft’s 2024 Work Trend Index, based on a survey of 31,000 knowledge workers in 31 markets, 78% of AI users said they were bringing their own AI tools to work, rising to 80% at small and medium-sized companies. When that happens without guidance, three problems follow:
- Data leaves your control. Customer records, financial figures, source code or confidential contracts get pasted into tools whose terms your business has never reviewed.
- Quality and accountability blur. AI-generated content reaches customers without review, and nobody is sure who is responsible when it is wrong.
- Legal exposure grows quietly. Existing laws on consumer protection, discrimination, privacy and intellectual property apply to AI-assisted work just as they apply to any other work.
The US Federal Trade Commission made the last point explicit when it announced Operation AI Comply in September 2024, stating that “there is no AI exemption from the laws on the books.”
A policy does not slow adoption down. Done well, it speeds it up, because employees know what they are allowed to do and new tools can be approved through a known process instead of case-by-case debate.
AI policy vs. AI governance framework
The two terms are often used interchangeably, but they do different jobs.
| AI use policy | AI governance framework | |
|---|---|---|
| Purpose | Tells people what they may and may not do with AI | Defines how the organization decides, oversees and improves its use of AI |
| Audience | Every employee and contractor | Leadership, system owners, IT, legal and risk |
| Typical length | A few pages, written in plain language | A set of roles, processes, registers and review cycles |
| Examples of content | Approved tools, prohibited data, review and disclosure rules | Risk tiers, approval workflow, AI inventory, monitoring, incident handling |
| Changes | Updated when rules change | Runs continuously as new uses are proposed and reviewed |
A small business can start with the policy and a minimal framework: one owner, one register and one approval route. The framework grows as the number and risk of AI uses grows.
The NIST AI Risk Management Framework in plain English
The NIST AI Risk Management Framework (AI RMF), released on January 26, 2023, is the most widely referenced US framework for managing AI risk. It is intended for voluntary use, so adopting it is a choice, not a legal requirement. NIST also notes that the AI RMF 1.0 is being revised as part of the White House AI Action Plan, so it is worth checking for updates when you adopt it.
The framework document organizes AI risk management into four core functions:
| Function | What it means | What it looks like in a smaller business |
|---|---|---|
| Govern | Culture, policies, roles and accountability for AI risk | A named AI owner, the written policy and an approval process |
| Map | Understanding the context of each AI use and its potential impacts | An AI inventory recording what each tool does, for whom and with what data |
| Measure | Assessing and tracking risks and trustworthiness | Testing outputs before launch and spot-checking quality afterward |
| Manage | Prioritizing and acting on risks, including incidents | Fixing or retiring problem uses and learning from incidents |
NIST describes Govern as “a cross-cutting function that is infused throughout AI risk management and enables the other functions.” In practice, that means your policy and ownership come first, and everything else depends on them.
The framework also lists characteristics of trustworthy AI: valid and reliable; safe; secure and resilient; accountable and transparent; explainable and interpretable; privacy-enhanced; and fair with harmful bias managed. These make a useful checklist when reviewing any proposed AI use.
The Generative AI Profile
In July 2024, NIST published NIST AI 600-1, a companion profile for generative AI. It names 12 risks that are unique to or made worse by generative AI. Several map directly to everyday business use:
- Confabulation — confidently stated but false content, often called hallucination.
- Data privacy — leakage or misuse of personal information.
- Information security — new attack routes, including manipulation of AI systems.
- Intellectual property — questions around copyrighted or confidential material in inputs and outputs.
- Harmful bias or homogenization — outputs that disadvantage groups or flatten diversity of views.
- Human-AI configuration — problems in how people use and rely on AI, such as over-trusting its output.
- Value chain and component integration — risks inherited from third-party models, data and vendors.
You do not need to address all 12 in depth on day one. Use the list to make sure your policy has at least one clear rule for each risk that applies to your business.
Step 1: Assign ownership
Name one accountable executive for AI, often the COO, CTO or head of operations in a smaller company. Their job is not to approve every prompt, but to own the policy, the approval process and the AI inventory. For larger organizations, add a small cross-functional review group that includes IT or security, legal or compliance, and a representative from the business teams using AI.
Also decide who owns each AI system once it is in use. Every tool or workflow that touches customers, employees or sensitive data should have a named owner responsible for its quality, cost and incidents.
Step 2: Take an inventory of current AI use
You cannot govern what you cannot see. Run a short, non-punitive survey asking teams which AI tools they use, for what tasks and with what kinds of data. Check software subscriptions and expense reports, and ask your key software vendors which AI features are now switched on in the products you already pay for.
Record the results in a simple register. For each use, capture the tool, the owner, the purpose, the data involved, whether outputs reach customers or affect decisions about people, and where a human reviews the output. This register becomes the backbone of your governance framework.
Step 3: Classify uses by risk
Not every use of AI deserves the same scrutiny. A tiered approach keeps low-risk productivity uses fast while giving real attention to the uses that matter. A simple three-tier model works for most smaller organizations:
| Tier | Typical examples | Controls |
|---|---|---|
| Low | Drafting internal emails, brainstorming, summarizing public information | Approved tools only; no confidential data; user checks output |
| Medium | Customer-facing content, code generation, analysis of internal business data | Approved tools with business-grade data terms; mandatory human review before use; owner named in the register |
| High | Decisions affecting people’s employment, credit, housing, insurance, health or legal rights; autonomous actions in business systems | Formal review before launch; testing for accuracy and bias; documented human oversight; legal review; ongoing monitoring |
Some uses may belong in a fourth category: prohibited. Examples might include entering regulated personal data into consumer AI tools or using AI to make final decisions about individuals without human review. Where AI systems can take actions on their own, such as agents connected to your business software, treat them as at least medium risk. Our guide to AI agents for business explains the additional controls they need.
Work with GeekOzy
Put AI to work where it pays off.
Tell us what you're working on. We'll come back with honest thinking on a practical next step.
Start a projectExplore AI & AutomationStep 4: Write the acceptable use rules
This is the part every employee reads, so keep it short and concrete. A practical AI use policy usually covers the following.
Approved tools
List the AI tools that are approved, for which tiers of use, and under which accounts. Business or enterprise accounts often come with different data handling terms from free consumer versions, so specify which one employees must use. Explain how to request a new tool.
Data rules
Be explicit about what must never be entered into AI tools that are not approved for it. Typical categories include customer personal information, employee records, payment and financial account data, health information, passwords and credentials, unreleased financial results, source code and anything covered by a confidentiality agreement.
Human review and accountability
State that people remain responsible for work they produce with AI assistance. Require review of AI output before it is sent to customers, published, used in code that ships or used in decisions about people. Remind staff that AI can produce plausible but false information.
Transparency and disclosure
Decide when customers or other people should be told that they are interacting with AI or receiving AI-generated content. Some laws already require notice in specific situations, as the legal section below shows, so check this with counsel for your sector and states.
Intellectual property and confidentiality
Address the use of third-party content in prompts, ownership of AI-assisted work and the need to respect client confidentiality. Your contracts with clients may already restrict how their information can be processed.
Prohibited uses
List uses that are not allowed at all, such as creating misleading content, impersonating real people or making final automated decisions about individuals in high-risk areas.
A one-page policy outline
If you need a starting structure, the following sections cover the essentials:
- Purpose and scope — why the policy exists and who it applies to, including contractors.
- Ownership — the accountable executive and how to reach them.
- Approved tools — the current list and how to request additions.
- Data rules — categories that must never be entered into unapproved tools.
- Human review — when AI output must be checked and by whom.
- Disclosure — when to tell people that AI is involved.
- Prohibited uses — what is not allowed.
- Risk tiers and approval — how new uses are assessed.
- Incident reporting — how to report problems.
- Review cycle — when the policy will next be updated.
Step 5: Build a simple approval process
A good approval process is quick for low-risk requests and thorough for high-risk ones. For each new tool or use, the requester describes the purpose, data and users. The AI owner assigns a risk tier and routes it: low-risk requests can be approved in days, while high-risk uses go to the review group.
For vendor tools, ask a consistent set of questions:
- Is our data used to train the vendor’s models, and can that be turned off?
- Where is data stored and processed, and for how long is it retained?
- What security certifications or independent assessments does the vendor hold?
- Which third-party models or subprocessors does the product rely on?
- What logging and administrative controls are available to us?
- How will the vendor notify us of incidents or material changes to the product?
The answers belong in your register next to the tool.
Step 6: Monitor, train and improve
Monitoring and incidents
Governance is not complete when a tool is approved. For medium- and high-risk uses, define what “good” looks like and check it regularly: accuracy on a sample of outputs, complaints or corrections, cost and usage. Review the register at least a few times a year to remove unused tools and catch new ones.
Set up a clear route for reporting AI problems, such as a wrong answer given to a customer, a suspected data leak or biased output. Treat these like other operational incidents: contain, investigate, fix and record what changed. NIST’s Manage function covers exactly this loop.
Training people, not just publishing rules
A policy that nobody understands will not be followed. Give every employee a short briefing on the rules, with realistic examples from their own work. Give heavier users practical training on writing good prompts, checking outputs and recognizing the limits of the tools. Make it easy to ask questions, and treat early mistakes as a chance to improve the policy rather than a reason to punish people for being candid about their AI use.
What laws and standards should you know about?
AI regulation is changing quickly, and obligations depend on your industry, location, customers and how you use AI. The following are useful reference points, not a complete list, and you should consult legal counsel about how they apply to your business.
- Existing US law. Consumer protection, anti-discrimination, privacy and sector rules continue to apply when AI is involved, as the FTC’s Operation AI Comply announcement emphasized.
- State AI laws. Colorado’s governor signed SB26-189, on automated decision-making technology, on May 14, 2026. According to the Colorado Division of Real Estate’s summary, it applies to consequential decisions made on and after January 1, 2027, and requires deployers to give notice of its use and to provide information to consumers after adverse outcomes. Other states have their own rules, and the landscape continues to shift.
- The EU AI Act. If you offer AI-enabled products or services in the European Union, the European Commission’s AI Act page states that the Act entered into force on August 1, 2024, with prohibited practices applying from February 2, 2025. Following the Digital Omnibus changes, the Commission lists December 2, 2027 as the start date for obligations on high-risk systems in sensitive areas such as employment and education.
- ISO/IEC 42001. Published in December 2023, ISO/IEC 42001 specifies an AI management system for organizations that develop, provide or use AI. It is certifiable and may matter if enterprise customers ask about your AI practices.
For most small and mid-sized businesses, a policy built on the NIST AI RMF gives a sensible foundation that can be extended toward specific legal requirements or ISO/IEC 42001 as needed.
Common mistakes to avoid
- Banning AI outright. Blanket bans tend to push use out of sight rather than stop it.
- Copying a large company’s policy. A long, legalistic document written for a different organization will not be read or followed.
- Treating the policy as a one-time project. Tools, laws and uses change quickly, so schedule regular reviews.
- Ignoring AI inside existing software. Many tools you already use have added AI features that may process your data differently.
- Leaving ownership vague. Without a named owner, the register goes stale and approvals stall.
The bottom line
An AI use policy tells people how to use AI safely. A governance framework makes sure the policy is followed, measured and improved. Start with ownership, an inventory and a short set of clear rules, use the NIST AI RMF’s Govern, Map, Measure and Manage functions to structure what follows, and scale oversight to the risk of each use.
If you would like help turning this into a policy and governance model that fits your business, explore our Strategy & Consulting services and our AI Transformation approach, or start a conversation with our team.
Sources
- Work Trend Index 2024: AI at work is here. Now comes the hard part — Microsoft and LinkedIn
- FTC Announces Crackdown on Deceptive AI Claims and Schemes — Federal Trade Commission (September 2024)
- AI Risk Management Framework — NIST
- Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1
- AI RMF: Generative Artificial Intelligence Profile, NIST AI 600-1
- Governor Polis Signs Bills into Law (SB26-189) — Office of the Governor of Colorado (May 2026)
- SB26-189 Summary — Colorado Division of Real Estate
- AI Act: Regulatory framework for AI — European Commission
- ISO/IEC 42001:2023 Artificial intelligence — Management system — ISO



