AI Acceptable Use Policy is a security control mapped to NIST CSF 2.0 GV.PO-01 and CIS Control 14 (Security Awareness and Skills Training). It helps small businesses reduce breach risk, including from AI-enabled attacks, and answer the security questions that insurers and clients ask.
An AI acceptable use policy is a short, written rule for how your organization uses artificial intelligence (AI) tools: chatbots, writing assistants, meeting note-takers, and the AI features built into software you already use. It names the tools staff may use, lists the information that must never go into them, and says who approves new tools. It turns hundreds of individual judgment calls into one shared standard.
The main risk it addresses is simple. Anything typed or uploaded into an AI tool leaves your systems and is held by the provider under its own terms, and many personal and free accounts may use conversations to train future models unless that setting is turned off. AI assistants connected to email or files add a second risk: hidden instructions planted in a message or document can try to steer the assistant, a technique called prompt injection.
Start by knowing which AI tools are in use, including ones staff adopted on their own. The NIST AI Risk Management Framework (AI RMF 1.0, January 2023) calls for mechanisms to inventory AI systems (GOVERN 1.6). NIST's Generative AI Profile (NIST AI 600-1, July 2024) adds keeping approved lists of AI technology and service providers.
Next, keep sensitive data in approved tools on business accounts, and review AI features in existing software before they are switched on, as you would a new vendor. Then make the policy part of training. The AI RMF expects staff to receive AI risk training consistent with your policies (GOVERN 2.2), and CIS Controls v8.1 Safeguards 14.1 and 14.4 cover a security awareness program and training on data handling.
In NIST Cybersecurity Framework (CSF) 2.0 terms, this is policy (GV.PO-01): established, communicated, and enforced. Keep it to a page so people actually read it, and review it at least once a year or whenever you adopt a significant new AI tool, as GV.PO-02 expects for changes in technology and threats.
Some cyber insurers have started asking how you govern AI use, but the questions vary from one insurer to the next and are still settling. What insurers already ask about (where sensitive data is stored, which outside services handle it, and whether staff are trained) applies directly to AI tools. A written AI rule lets you answer those questions honestly.
AI also shows up in the incidents themselves. The FBI's 2025 Internet Crime Report counted more than 22,000 complaints that referenced AI, with reported losses above $893 million. A clear AI rule, paired with the callback habit in our Out-of-Band Verification control, covers both directions: the data you send to AI tools and the AI-made fakes sent to you.
Entering protected health information (PHI) into an AI tool discloses it to that tool's provider. Under HIPAA, a covered entity may let a vendor create, receive, maintain, or transmit PHI on its behalf only with satisfactory assurances documented in a written business associate agreement (45 CFR 164.502(e) and 164.504(e)). Many consumer AI accounts are not offered with one, so PHI belongs only in AI tools covered by a signed agreement.
The FTC Safeguards Rule requires covered financial institutions to identify and manage the data and systems that support the business, and to select, contract with, and periodically assess service providers (16 CFR 314.4(c)(2) and 314.4(f)). An AI provider that receives customer information through its service fits the Rule's definition of a service provider (16 CFR 314.2). FTC Safeguards Rule citation: https://www.ftc.gov/business-guidance/privacy-security/safeguards-rule
Card numbers should never be entered into an AI tool. Any system that stores, processes, or transmits cardholder data falls within PCI DSS scope, so a chatbot that receives card data pulls that service into your compliance obligations. Customer names, addresses, and order histories deserve the same care under state privacy and breach-notification laws.
Defense contractors must protect Controlled Unclassified Information (CUI) on systems that meet NIST SP 800-171 (DFARS 252.204-7012(b)(2)(i)). A cloud service that stores, processes, or transmits covered defense information must meet security requirements equivalent to the FedRAMP Moderate baseline (DFARS 252.204-7012(b)(2)(ii)(D)). A commercial AI chatbot that has not met that bar is not a place for CUI.
List every AI tool and AI feature staff use for work, including free chatbots, browser extensions, meeting note-takers, and assistants built into existing software
Write a one-page rule naming the approved tools, the information that must never go into AI tools, and who approves new tools
Move approved AI use to business accounts and turn off settings that let providers train on your data
Review AI features in existing software before enabling them, and require a person to confirm before an assistant sends, shares, or pays
Have every staff member acknowledge the rule, include it in onboarding, and review it at least once a year or when you adopt a significant new AI tool
NIST CSF 2.0, Govern: GV.PO-01
NIST Cybersecurity Framework 2.0CIS Control 14: Security Awareness and Skills Training
CIS ControlsAn AI acceptable use policy is a short, written rule for how your organization uses artificial intelligence (AI) tools: chatbots, writing assistants, meeting note-takers, and the AI features built into software you already use. It names the tools staff may use, lists the information that must never go into them, and says who approves new tools. It turns hundreds of individual judgment calls into one shared standard.
Some cyber insurers have started asking how you govern AI use, but the questions vary from one insurer to the next and are still settling. What insurers already ask about (where sensitive data is stored, which outside services handle it, and whether staff are trained) applies directly to AI tools. A written AI rule lets you answer those questions honestly.
List every AI tool and AI feature staff use for work, including free chatbots, browser extensions, meeting note-takers, and assistants built into existing software Then, Write a one-page rule naming the approved tools, the information that must never go into AI tools, and who approves new tools Then, Move approved AI use to business accounts and turn off settings that let providers train on your data Then, Review AI features in existing software before enabling them, and require a person to confirm before an assistant sends, shares, or pays Then, Have every staff member acknowledge the rule, include it in onboarding, and review it at least once a year or when you adopt a significant new AI tool.
See how this control stops real-world attacks, including AI-enabled scams:
Want to know how your organization measures up on this control?
Start the free Quick Check