Blog
Benchmarks, methodology, engineering notes.
Where promptShield fits: a map of how companies protect PII — and the gap we fill
Companies protect personal data in four broad ways — governance platforms, cloud redaction APIs, managed redaction, and manual blackout. Each is built for a different shape of problem. Here's the honest map, and the specific gap promptShield is built for: a small number of high-value documents that one person has to be able to vouch for.
Read the postWhy we built our own PII engine instead of shipping Presidio
Presidio is an excellent toolkit. We still wrote our own detection engine — because the thing we ship is a bounded PDF on someone's laptop, not a text stream in a cloud pipeline. The architectural case, honestly told.
Read the postIs it safe to paste documents into ChatGPT?
Short answer: sometimes — and which case you're in is entirely under your control. A practical guide to what determines the risk, what actually happens to the text you paste, and the one step that makes the question moot.
Read the postTo those who quietly share client documents with a chatbot
AI tools make confidential data exposure effortless — and the consequences are neither theoretical nor distant. Here is what knowledge workers are actually risking when they paste client documents into ChatGPT, Claude, or Gemini.
Read the postPII protection looks completely different when AI is in the loop
For 30 years, protecting personal data meant controlling who could reach the database. AI tools broke that model by creating a frictionless path from controlled documents to external cloud processing. Here is what the new approach requires.
Read the postWe benchmarked our PII detection against Microsoft Presidio — here's what we learned
A reproducible, head-to-head benchmark on 14 PDF documents across 7 European languages. Microsoft Presidio's default install emits 666 spans; promptShield emits 252 on the same corpus. Most of the gap isn't capability — it's noise we filter that Presidio doesn't.
Read the post
