AI vendor terms: four clauses to check before you adopt a tool

The demo shows what the tool can do; the terms show what the vendor may do, and customer data lives in the gap between them.

Adopting an AI tool takes an afternoon: card details, a workspace, the whole team invited by evening. The part that gets skipped is the subscription agreement, and it is the only part written by the other side. Inside it sit clauses that decide what happens to your material: the proposals you draft, the client documents you summarize, the history your team accumulates inside the tool.

The honest framing is a gap. The demo shows what the tool can do; the terms show what the vendor may do. Nothing in the product tour contradicts a clause that lets your inputs improve a shared model, because the tour is not the contract. Between the two sits the quiet transaction most businesses never notice they agreed to.

The failure versions are unremarkable, which is what makes them common. Business documents fed into a tool whose terms allowed those inputs to train a model shared with strangers. A customer conversation stored on servers in a jurisdiction nobody checked, under rules nobody read. A cancellation executed cleanly, which is when the team discovers there is no export, and the accumulated work history evaporates with the login.

Before adopting, check four things in the terms. Whether your inputs train anything shared, and whether a setting exists to switch that off. Where the data lives, which country's servers hold it, and who inside the vendor can reach it. What happens on cancellation, specifically whether your history leaves with you in a form a human can open. And what the vendor may claim: rights over your content that outlive the subscription are not a detail.

Then prefer the configurations where your data stays your own. Most serious tools now ship a workspace setting that excludes customer content from shared training, and most businesses never switch it on, because the setting lives in an admin menu nobody owns. Treat the terms check the way you would treat credit terms from a supplier: routine, documented, and done before the relationship starts, not during a dispute.

We run AI through our own funnel, so these questions get asked of every tool we adopt, and we built our own assistant precisely so the answers stay ours. This essay is not legal advice: a hard reading of cross-border data terms belongs with licensed practitioners, and jurisdictions differ. Guidance, not guarantees: four checks before the card details, and the material stays yours.