An AI governance checklist for small businesses

Most small businesses did not decide to adopt AI; it arrived. A developer started using a coding assistant, the support team pasted tickets into a chatbot, marketing switched on the AI features in their email tool, and someone connected an automation that sends customer data to an LLM API. None of this is wrong, but without a few simple rules it is easy to leak customer data, publish something false, or discover a large bill at the end of the month.

Governance sounds like a large-company word. For a team of five to fifty people it simply means knowing which AI tools you use, what data goes into them, who checks their output, and what the law expects. This guide gives you a practical checklist built on public frameworks: the NIST AI Risk Management Framework, the OWASP Top 10 for LLM Applications, the EU AI Act timeline and India's Digital Personal Data Protection (DPDP) Rules. It is not legal advice; use it to organise the conversation with your adviser.

All dates and references were checked against official pages on 7 October 2026.

Why bother now

• Data protection law already applies. Sending personal data to an AI provider is processing like any other. In India, the government notified the DPDP Rules, 2025 on 14 November 2025 with an eighteen-month phased compliance period, so the main obligations are landing in 2027. Organisations serving EU residents already work under the GDPR.
• The EU AI Act is in force. According to the European Commission it entered into force on 1 August 2024 and became applicable on 2 August 2026, with some exceptions. If you sell into the EU, parts of it can apply to you even if you are based elsewhere.
• Customers ask. Security questionnaires from larger clients increasingly include questions about AI use and data handling.

A simple framework: Govern, Map, Measure, Manage

NIST released its AI Risk Management Framework (AI RMF 1.0) on 26 January 2023 and a Generative AI Profile (NIST AI 600-1) on 26 July 2024. It is voluntary and free, and its four functions make a useful skeleton even for a very small company:

Function	What it means for a small team
Govern	Name an owner, write a one-page AI use policy, decide who approves new tools.
Map	List every AI tool and use case, with the data each one touches.
Measure	Check output quality, cost and incidents on a regular schedule.
Manage	Act on what you find: restrict, retrain, switch vendor or stop a use case.

NIST notes that the AI RMF is being revised as part of the US AI Action Plan, so watch for a new version.

The checklist

1. Ownership and policy

• Name one person accountable for AI use (often the CTO, operations lead or founder).
• Write a one-page policy covering: approved tools, data that must never be entered, when human review is required, and how to report a problem.
• Decide how new AI tools get approved, including AI features switched on inside software you already pay for.

2. Inventory

• List each tool, its plan tier, who uses it, the purpose, and what data goes in.
• Note whether it is a consumer account or a business account. Business tiers often have different data-retention and training terms; check the vendor's terms for your exact plan.
• Include API integrations and automations, not just chat tools.

3. Data rules

• Classify data simply: public, internal, confidential, personal, sensitive personal.
• Default rule: no confidential or personal data in tools that are not approved for it.
• Check where the vendor processes and stores data, how long it retains prompts and outputs, and whether it uses them for training.
• Update your privacy notice if AI tools process customer personal data.

4. Vendor review

• Business plan with admin console, single sign-on and the ability to remove users.
• Security documentation (for example a SOC 2 report or ISO certifications) and a data processing agreement.
• Spend controls: budgets, alerts and per-user limits.
• An exit plan: can you export your data and switch provider?

5. Human review of output

• AI-generated code goes through the same review, tests and security scanning as any other code.
• Anything published (articles, product descriptions, legal or medical statements) is checked by a person before it goes out.
• Decisions about people (hiring, credit, access to services) are not made by AI alone. In the EU, employment is one of the high-risk areas.
• Label AI-generated content where your audience would reasonably expect to know.

6. Security for AI features you build

If your product calls an LLM, review it against the OWASP Top 10 for LLM Applications (2025 edition). The list is:

1. LLM01 Prompt Injection
2. LLM02 Sensitive Information Disclosure
3. LLM03 Supply Chain
4. LLM04 Data and Model Poisoning
5. LLM05 Improper Output Handling
6. LLM06 Excessive Agency
7. LLM07 System Prompt Leakage
8. LLM08 Vector and Embedding Weaknesses
9. LLM09 Misinformation
10. LLM10 Unbounded Consumption

The practical controls for a small team: treat retrieved content and tool results as untrusted input; never let the model's output run as code or SQL without validation; give AI agents the narrowest permissions and require confirmation for actions that spend money, send messages or delete data; keep API keys on the server; and cap usage per user so a bug or an attacker cannot run up costs.

7. Logging and incidents

• Log which AI features were used, by whom and at what cost; keep prompts and outputs only as long as you need them, under the same access controls as the underlying data.
• Add AI incidents (data pasted into the wrong tool, harmful or false output published, runaway cost) to your normal incident process.
• Know your breach notification duties. India's DPDP Rules set out a process for notifying the Data Protection Board and affected individuals of a personal data breach.

8. Training

• Run a short session for everyone on the policy: what is allowed, what is not, and why.
• The EU AI Act's AI literacy obligation has applied since 2 February 2025, alongside the bans on prohibited practices. If the Act applies to you, keep a record of the training you provide.

Regulatory dates to keep on your calendar

Date	What applies	Source
2 February 2025	EU AI Act: prohibited practices and AI literacy	European Commission
2 August 2025	EU AI Act: obligations for general-purpose AI models; governance rules	European Commission
14 November 2025	India: DPDP Rules, 2025 notified; eighteen-month phased compliance begins	Press Information Bureau
2 August 2026	EU AI Act generally applicable, including Article 50 transparency obligations	European Commission
2 December 2027	EU AI Act: high-risk systems in sensitive areas (such as biometrics, education, employment)	European Commission
2 August 2028	EU AI Act: high-risk AI in regulated products (such as lifts or toys)	European Commission

The two high-risk dates reflect the "Digital Omnibus" amendments, which the Commission says entered into force on 27 July 2026 and staggered the original timeline. Most small businesses using AI for coding, content or support will not be running high-risk systems, but the transparency rules (for example, telling people they are interacting with an AI system, or labelling certain AI-generated content) are more likely to touch customer-facing chatbots and content tools. Check Article 50 with your adviser if you serve EU users.

A 30-day starter plan

1. Week 1: name the owner, send a short survey asking every team member which AI tools they use, and build the inventory.
2. Week 2: classify the data each tool touches, move important use cases to business accounts, remove unapproved tools that handle personal data.
3. Week 3: write and share the one-page policy; set budgets and alerts on every paid AI service.
4. Week 4: run the training session, add AI to your incident process, and book a quarterly review.

A template for the one-page policy

Keep it short enough that people actually read it. A workable structure:

1. Purpose. "We use AI tools to work faster. This policy keeps customer data safe and our output accurate."
2. Approved tools. A list with the account type for each (for example, "coding assistant: business plan only, via company SSO").
3. Never enter. Passwords, API keys, customer personal data, payment details, unreleased financials and anything under NDA, unless the tool is approved for that data class.
4. Always review. Code, published content, customer-facing replies and anything with legal, financial or medical meaning.
5. Disclose. When and how we tell customers that AI was involved.
6. New tools. Ask the owner first; include AI features inside existing software.
7. Problems. How to report a mistake (for example, data pasted into the wrong tool) without blame, and how fast.
8. Review date. The policy is reviewed every quarter.

Summary

AI governance for a small business is an inventory, a one-page policy, clear data rules, human review where it matters, basic security for anything you build, and a calendar of legal dates. Start with what you already use, keep the documents short enough that people read them, and review them every quarter as tools and rules change.