An email address can end up in an AI prompt without anyone meaning to share it. Someone pastes a customer message, a support ticket, or part of a document, and the address travels with the rest of the text.I built a Chrome and Firefox extension that catches detected email addresses in ChatGPT requests and replaces them with [EMAIL_ADDRESS] before the request leaves the browser.The important design decision wasn’t the placeholder. It was where the masking happens. Redacting the text after it reaches the service is already too late.The scope is deliberately narrow: detect email addresses in this browser workflow, replace them locally, and leave the surrounding prompt intact. Narrow privacy controls are easier to explain, test, and trust than broad claims about “protecting all PII.”Where would you put this kind of check in your own AI workflow: the browser, an API gateway, or both?