Start with the failure, not the AI label
AI can make fraudulent messages, voice requests and fake documents more convincing or easier to produce. A small business still needs to protect the same critical decisions: who can sign in, who can change payment details, who can approve a transfer and who can restore work after an account is compromised. Do not buy an “AI threat” product before proving that those workflows are controlled.
NIST SP 1300 is a final, small-business quick-start guide for using Cybersecurity Framework 2.0. It is a practical structure for identifying assets, protecting access, detecting problems, responding and recovering. NIST's AI Risk Management Framework resources address risks from AI systems your business builds or uses. They are related, but neither publication certifies your organization or gives a universal “AI-proof” checklist.
Three controls to implement this month
1. Verify payment changes outside the message
A familiar-looking email, chat, meeting transcript or voice request is not proof of authority. When a supplier changes bank details or an executive requests an unusual transfer, call a trusted number already in your records, not a number in the request. Require a second approver for significant transfers. Record the exception and who approved it. Test the workflow with a harmless simulated request.
2. Protect the accounts that can act
Require phishing-resistant MFA where practical for admins and payment approvers; otherwise use the strongest MFA available and train users to reject unexpected prompts. Remove former staff promptly, review privileged roles and keep recovery methods current. An AI-generated message is far less useful to an attacker if it cannot persuade a user to grant access or approve an action.
3. Make response and recovery usable
Write down how staff report a suspicious request, who pauses a payment and who contacts the bank or vendor. Keep tested backups of critical data and exportable records. Practice an account-lockout and a representative restore. A detection tool is less valuable than a team that knows what to do in the first hour.
A 30/60/90-day route
| When | Action | Evidence it works |
|---|---|---|
| First 30 days | Map payment and admin-account workflows; add independent verification and MFA | Two owners can demonstrate a blocked test request and account recovery |
| By day 60 | Review vendor changes, shared access and recovery copies | Sample vendor change has second-party approval; representative file restore succeeds |
| By day 90 | Run a short impersonation exercise and fix the failure points | Staff report through the intended route; exceptions are assigned and closed |
Use the free security assessment to identify whether identity, backup or response is currently your largest gap. Unknown answers are a reason to verify, not evidence of protection. If you deploy generative AI internally, inventory where company or customer data is entered and evaluate that use under the NIST Generative AI Profile. Apply procurement, access and data-retention rules to AI tools as you would to any other service.
What this guidance does not claim
This page does not estimate how often AI fraud affects your industry, promise that a control will stop every attempt, or claim NIST compliance. It is editorial interpretation of official NIST guidance reviewed September 25, 2026, not a certification or legal opinion. See our methodology and disclosure.
Turn concern into three actions
Use the free assessment to prioritize controls you can actually verify.
Start the free assessment