Notes on privacy, security and running AI inside a small firm or practice.
The honest, specific answer to the question every practice asks before adopting an AI tool. Not "your data is safe," but exactly what goes out, what stays, and what happens if the safeguard fails.
A share button, a public URL and a search crawler were all it took. No breach, no attacker. Here is why that particular failure has no equivalent on a device that lives in your own office.
When we wrote about the Claude share-link leak, we said the real fix isn't caution, it's auditing. So we ran that same audit on ourselves. It worked exactly the way it's supposed to.