Twenty minutes on the thing nobody in your organisation has measured.

While leadership evaluates enterprise AI platforms, your engineers have been pasting stack traces, config files, proprietary logic and occasionally customer data into whatever consumer chatbot is open in the next tab. Not out of recklessness — because the sanctioned option is slow, restricted, or doesn't exist yet. Every week without a supported pathway makes the workaround more entrenched.

This talk is about naming it and fixing it, without pretending a ban will work.

Why prohibition reliably fails, and what a sanctioned pathway needs to offer instead
The four categories of data that leak most often, and the one that will actually end up in an incident report
Why the biggest exposure in an enterprise AI rollout isn't the model — it's file permissions, and an assistant that faithfully surfaces everything a user was already over-entitled to see
A one-page approved-tools register and data rule you can publish next week
How to have this conversation with your team without it sounding like an accusation

Short, uncomfortable in the useful way, and immediately actionable.