Run through this before anything leaves your desk: a message to a customer, a number in a report, a decision about a person. It’s organized around the three moments that matter most.
Before you use an AI tool
- Read what an AI tool wants access to. Grant only the access the task needs. T-007 →
- Connect one account at a time. Start with a single low-risk account or folder, then expand. T-008 →
- Start with low-risk tasks. Drafts, brainstorms and summaries of public information first. T-011 →
- Write a one-page AI rule. Which tools are approved, what never goes in, who reviews what. T-012 →
- Name an owner for every AI tool. One person tracks what it can do and who uses it. T-013 →
Before you paste information in
- Keep customer data out of public AI tools. Names, emails and order details stay out of tools you haven’t approved. T-001 →
- Keep contracts, pricing and HR notes private. That information carries legal and competitive weight. T-002 →
- Ask who can see it before you send. Know who can see a prompt and how long it’s kept. T-003 →
Before you send or publish
- Treat every AI answer as a draft. Fluent isn’t the same as correct. T-004 →
- Check every detail against the original. Go back to the source before it goes into a document or decision. T-005 →
- Open every link the AI cites. Confirm the source exists and says what it’s credited with saying. T-006 →
- Read it yourself before it goes out. No exceptions for customer emails, public posts or proposals. T-009 →
- Keep people and money decisions human. Hiring, firing, pricing and payments stay with a person. T-010 →
- Approve any AI action you can’t undo. Sending, paying, deleting and publishing wait for a person. T-014 →
- Tell people when they’re dealing with AI. Label AI replies and offer an easy way to reach a person. T-016 →
Sources and further reading
Built from our AI Safety Tips. Checked against the NIST AI Risk Management Framework and the U.S. Small Business Administration’s guidance on AI for small business.
