PII ShieldBlog
Keeping personal data out of AI, in practice.
One piece per surface. Each answers a question somebody actually asks, in the first paragraph, before the depth.
How to connect your ATS or CRM to Claude without exposing personal data
Adding your ATS to Claude as a connector does not protect it — Claude talks to it directly and every name it receives is real. Here is what to put in between, and why one address for several systems matters.
How to stop staff pasting customer data into ChatGPT and Claude
Policy does not survive contact with a deadline. What works is masking the text in the chat box before it is sent — and restoring the real values in the reply, so nobody has a reason to route around it.
How to redact PII before it reaches the OpenAI or Anthropic API
A gateway is a base-URL change: redaction happens on every request whether or not anyone remembered, which is the only version that holds when the codebase grows.
How to redact a Word, Excel or PDF file before uploading it to AI
Deleting text from a PDF usually does not delete it. What format-preserving redaction has to get right, and why scanned pages need OCR before anything can be masked at all.
How to find out what personal data has already gone into your team's AI chats
Prevention says nothing about the last eighteen months. Auditing existing chat history is the uncomfortable half — including shared links, which are public web pages.