The idea
AI risk is often discussed in dramatic terms. For most businesses the real risks are ordinary and manageable. They come down to three questions about information and one about responsibility.
What goes in
Anything typed or uploaded into an AI tool is sent somewhere. For each tool your people use, you should know where the information is processed, how long it is kept, and whether it is used to train the provider's models. The answers are in the provider's terms and differ between free and business versions of the same product.
Decide which kinds of information must never go into which tools: customer data, personal data, financials, and anything under a confidentiality agreement.
What comes out
AI output can be wrong and sound certain. Decide where a person must review it before it is used. A draft for internal use needs little checking. Anything sent to a customer, entered into a record, or used to make a decision about a person needs more.
Who can see what
An assistant connected to your documents should not show a person anything they could not have opened themselves. Access rules that already exist need to carry through to the AI system. This is a design requirement, and it is much easier to build in than to add.
Who is responsible
Every use of AI needs a person who answers for the result. "The AI decided" is not an answer your customers, your staff, or a regulator will accept.
Depending on your industry and where you operate, specific rules and frameworks may apply, such as HIPAA, the NIST AI Risk Management Framework, or the EU AI Act. This lesson is general guidance, not legal advice. Check your own obligations with someone qualified.
A usage policy on one page
- Which tools are approved, and for what.
- What information must not be entered.
- When a person must review output before it is used.
- Who to ask when it is unclear.
- How to report a mistake or a concern.
A short policy people read is worth more than a long one they do not.
Exercise · about 15 minutes
Using the five headings above, write one or two lines under each for your own team. You will have the outline of a usage policy.
One thing to take to your team
Find out which AI tools your people already use. The honest answer is usually more than the approved list.