Someone on your team used an AI tool today. Maybe it drafted an email. Maybe it summarized a contract or cleaned up a spreadsheet. They’re trying to save time. Without a policy, they’re guessing which tools are safe and what’s off limits.

An AI policy closes that gap, and it works best in two layers. Layer one is a one-page staff guide that people use day to day. Layer two is the full policy behind it, which is what auditors, insurers, and client security questionnaires ask to see. Here’s what goes in each policy, and how to get both into your team’s heads.

Your staff is already ahead of you

Shadow AI is what happens when people use AI tools nobody approved. It isn’t rare. IBM’s 2026 Cost of a Data Breach Report found shadow AI involved in 43% of the security incidents it studied, up from 20% the year before. Those incidents averaged $5.39 million, against a $4.99 million average for all breaches. IBM also found that 68% of the organizations it looked at lacked AI governance policies.

Most people aren’t being reckless. They’re trying to save time, and nobody told them where the lines are. A clear page fixes that.

Layer one: the one-page staff guide

A one-page guide is easy to find and easy to remember. It also forces you to decide what matters. If a rule can’t fit on the page, people are unlikely to recall it at 4:45 PM on a Friday. Anything that needs more detail belongs in layer two.

What goes on the page

  1. Purpose and scope. Two sentences. Say who it covers: employees, contractors, and interns. Say what it covers: any AI tool, including AI features inside software you already use, meeting note-takers, browser extensions, and personal accounts used for work.
  2. Approved tools. Name them. List which tools staff can use for work and require the company login every time. Anything else needs approval first. Our tool-choice post can help you build the list.
  3. Data rules. This is the heart of the page. Sort your data into three tiers (see below).
  4. Human review. AI drafts. People decide. Anyone who sends AI-assisted work to a client, regulator, or the public checks the facts, numbers, and sources first. You own what you send.
  5. Approval required. Some uses need a yes first: recording or transcribing meetings, hiring and HR decisions, anything customer-facing, and anything that touches regulated data. Recording consent rules vary by state.
  6. Reporting. If someone enters something they shouldn’t, they tell IT right away. No blame. A fast report is the best outcome, and it feeds straight into your incident response plan.
  7. Owner and review date. Name one person who owns the page and approves new tools. Review it every six months. AI tools change faster than most policies.

The three tiers

TierExamplesRule
OpenPublished marketing copy, public information, general how-to questionsOkay in approved tools
InternalDraft documents, meeting notes, internal procedures, non-sensitive project detailsApproved tools only, with your company login
NeverCustomer, patient, or employee data. Financial records. Passwords and keys. IP addresses, server names, network diagrams, and logs. Regulated data such as HIPAA, payment card, or CUI.Never in unapproved tools. Only in tools IT has set up for that kind of data

Note: Some items are a hard no in every tool, approved or not: passwords, keys, and anything under NDA.

What to tell staff

The page is the reference. The conversation is what changes behavior. Shorten the message down to five lines:

  • Use approved tools, with your work login. Personal accounts for work are off limits.
  • When in doubt, leave it out. Describe the situation instead of pasting the details. Write “a client disputing an invoice” instead of pasting the invoice.
  • AI drafts. You decide. Check everything before it leaves your hands.
  • Ask before you try something new. A new tool, a new use, a new plugin. Expect a quick yes or no.
  • Tell us if you slip. Nobody gets in trouble for reporting a mistake fast. Hiding one costs more.

Then make it stick. Run a 15-minute huddle with real examples from your own business. Post the page where people can find it. Ask managers to repeat it. Add it to onboarding and have people sign the acknowledgment described below.

Also answer “can I use this?” fast. Name a person or a shared mailbox, and commit to a reply within two business days. A slow yes is how shadow AI starts.

Layer two: the full policy behind the page

The one-pager tells staff what to do. The full policy records how your company governs AI. It’s the document a prospect, auditor, or insurance carrier will ask about. It doesn’t need to be long or complicated, but it does cover things that won’t fit on a single page. You don’t need to draft it all at once. Here’s what lives there:

  • Accountability. Who owns AI governance, who signs off on exceptions, and who answers to leadership.
  • Tool approval. How a tool gets reviewed and approved, what you check before saying yes, and how the approved list is maintained.
  • Vendors and third-party AI. How you assess the AI built into vendors’ products, and what you expect from the vendors who handle your data.
  • AI agents and automation. What tools that take actions on their own are allowed to access and do, and who approves them.
  • Data retention and training. How long AI tools keep your data, and whether they’re allowed to train on it.
  • Intellectual property. Who owns AI-assisted work, and how you protect your own and others’ IP.
  • Incident response. What happens after someone reports an AI-related mistake, and how it connects to your broader incident response plan.
  • Enforcement. What happens when the policy isn’t followed, applied consistently.
  • Review cycle. How often the policy is reviewed, who reviews it, and how changes are tracked.

Think of it this way: the one-pager is the daily-use summary, and the full policy is the source of truth behind it. When the two are built together, the page stays short because the detail has somewhere to live.

Tie the layers together with one acknowledgment

A yearly acknowledgment works best when people know exactly what they’re signing. Use a single acknowledgment that covers both documents, with a version number for each. By signing, each person confirms that they:

  • Have read the one-page staff guide.
  • Have access to the full policy.
  • Will follow both.
  • Understand that the full policy takes precedence if the two ever differ.

Have all staff sign again every year and whenever either document changes. Include contractors and interns. Version numbers let you show an auditor or insurer exactly which wording each person agreed to, and when.

Before you roll it out, have an employment attorney review the final wording.

Back the page with controls

A policy people ignore isn’t a control. IBM found that only about two in five organizations apply access controls to their AI models and data. Pair the page with technical guardrails.

Start with sign-in. Require your company login, with multifactor authentication, for every approved tool. Then limit the rest. Block or restrict unapproved AI sites on work devices, so the easy path is also the approved one.

Next, check what your AI tools can reach. Review their access in Microsoft 365, SharePoint, and shared drives. If a person can open a file, an AI assistant can usually surface it. Finally, turn on logging, so you can answer “who used what” later.

Five mistakes to avoid

  • Banning everything. People keep using AI on their phones, and you lose all visibility.
  • Only writing a one-page policy. The one-page policy is a great tool for staff, but it should not stop there. Auditors, insurers, and client questionnaires will want a policy behind it.
  • Hard-coding tool names in the policy. Keep the approved list on a separate sheet you can update.
  • Forgetting AI built into tools you already own. Email, meetings, CRM, and browsers all have it now.
  • No owner and no review date. A policy nobody maintains is out of date in months.

Your next steps

  1. Pick an AI owner. Ask your team which AI tools they already use, with no penalty for answering.
  2. Draft the one-page guide using the seven parts above.
  3. Outline the full policy using the nine topics in layer two.
  4. Create one acknowledgment that covers both documents, with version numbers, and have an employment attorney review the final wording.
  5. Get leadership sign-off, then brief your team within 30 days and collect signed acknowledgments, including from contractors and interns.
  6. Put a six-month review on the calendar, and repeat the acknowledgment each year and whenever either document changes.
  7. Check your client contracts and cyber insurance application. Some ask about AI use, and your answers should match your policy.

ADNET can help. We’ll review your draft, build the approved-tool list with you, set up the controls behind it, and help brief your team. Reach out at Engagement@thinkadnet.com.

ADNET Technologies | Engagement@thinkadnet.com

Rocky Hill, CT 860.409.1700  |  Albany, NY 518.458.9300