AI Is Already on Your Team, Whether You Budgeted for It or Not

Someone on your staff used AI for work this week. The question is not whether; the question is whose account they used, and whether they used it within the scope of regulations and policies governing your organization. 

That is not a guess. Cyberhaven Labs tracked billions of real data movements across 222 companies for its 2026 AI Adoption & Risk Report. Roughly a third of all ChatGPT usage runs through personal accounts. Nearly 40% of everything employees put into AI tools contains sensitive data, and the average employee does it once every three days. That is telemetry, not a survey; it is what people actually did, not what they said they did. 

Our own AI in Recreation Study points the same way from the other end. 80% of recreation professionals say they are familiar with AI, while close to 40% say their organization is not using it at all. Familiarity at that level does not come from nowhere. It is being built somewhere, on accounts nobody chose and nobody administers.

IT has a name for that: shadow IT. 

The cost you are avoiding is smaller than the cost you are already carrying

Nobody drafting a parent email in a personal ChatGPT account is trying to cause an incident. They are solving a problem the fastest way available to them, because the organization never gave them a sanctioned option. When the approved toolset contains no AI, staff supply their own. The gap is not behavior; it is procurement.

Today some organizations are paying nothing for AI and carrying the entire risk of it: member records, enrollments for minors, health notes, and payment details flowing into accounts they do not administer, cannot audit, and cannot revoke when that person resigns. The line item they did not create is far smaller than the exposure they already own.

There are three steps that will get governance sorted. 

First, write the policy

Policy comes before tooling, because a license without rules just moves the same behavior onto a bill you pay. You do not need a consultant to produce the first draft. Open an AI tool in research mode and give it this:

The short version is the one that changes behavior: which tools are approved, what must never be pasted into them, and who to ask when someone is unsure. In the meantime, counsel can review the thorough version people read and forget. 

Second, extend security awareness to prompt injection

Your staff have been trained for years to spot a suspicious email. Nobody has trained them on the worst version of that attack that targets LLMs.

Prompt injection is instructions hidden inside content the AI reads: a PDF, a web page, a message buried in the thread you asked it to summarize. The model follows them, because nothing in what it receives marks which words are your request and which are someone else’s order. OWASP has ranked it the number one security risk for LLM applications in every edition since 2023.

Your awareness program covers phishing well. It does not yet cover this one nearly as well, and it should. Add it to the same rotation, with the same cadence, in plain language. 

Third, buy the license that gives you governance

A business or enterprise account is not a fancier version of the same chatbot. It changes who is in control. Your prompts are contractually excluded from model training. An admin console gives you visibility into usage. Access can be reassigned or revoked when someone leaves, and the history stays with the organization instead of walking out the door with them.

For nonprofits, the price objection is weaker than most boards assume. Verified 501(c)(3) organizations get up to 75% off from Anthropic, Google, and OpenAI. Microsoft’s nonprofit discount on Copilot is considerably smaller. This gives you governed, enterprise-grade seats around $8 per user per month, less than half the retail price of the personal plan someone on your team may already be paying for out of pocket. All three verify eligibility through a third party, and the paperwork is the kind of thing a finance coordinator clears in an afternoon.

The line has moved

Amilia’s position has not changed in fifteen years: technology should take the work off your staff so their hours go where humans genuinely matter, which is the experience of your participants, the quality of your offer, and your reach in the community. AI moves that line further than any tool before it. 

To move with it, you have to unlock it for your organization first.