Verify before you paste

One workplace habit that changes how you use models this week - check what leaves your screen.

Most AI mistakes at work do not start with a bad model. They start with a fast paste.

Before you drop text into a chat, ask three checks out loud:

Would I email this to a stranger? If the paste has customer names, tickets, passwords, health details, or internal pricing, stop. Redact or use an approved tool.
What happens if a number is wrong? Titles and brainstorms are cheap. Invoices, policies, and status reports are not. Mark anything high-risk for a human check.
Can I point to a source? If the model invents a citation, a law, or a “company policy,” you own the error the moment you forward it.

This changes how you should work with a model this week: treat paste as a privilege, not a reflex. A skeptic would open your chat history and ask which messages contain secrets or unverified claims. You can say it in a few hundred words without a signup wall - because the fix is a habit, not a product.

What changed versus “just be careful”: make the check explicit at the moment of paste, not after the damage. Save one redacted example that worked. Refuse one paste that should never have left your machine.