Every enterprise leader we talk to has a rollout plan for the AI tools they've approved. Almost none of them have a plan for the AI tools their employees are already using without approval.

That's shadow AI, and it's usually the single biggest exposure channel in the whole stack — not because the tools themselves are dangerous, but because the terms are different. The consumer version of a tool your company already trusts under an enterprise agreement can have a completely different data-retention and training policy the moment an employee signs in with a personal account.

The pattern repeats everywhere: a team under deadline pressure finds the fastest path to an answer, pastes in something they shouldn't have, and nobody finds out until it shows up somewhere it shouldn't.

Shadow AI is usually the biggest exposure channel. Same tools, different contract, different data terms.

The fix isn't a ban — bans just push the behavior further out of sight. It's governed provisioning: give people a sanctioned path that's as fast as the unsanctioned one, so there's no reason to go around it. That's a platform hub decision, not a policy memo.