Strip PII out of your LLM logs

A customer typed their card number straight into a chat box. It's sitting in plain text in our logs, and it's not the only one.

Intermediate to advancedRead 7 minDoing: an afternoon6 steps

▶ Start

What you'll have at the end

A logging pipeline that masks names, cards, and emails before a single byte reaches disk

Before you start

You need

You already have an LLM observability or tracing tool wired into your app, capturing spans for your prompts, completions, and any tool calls it makes.

Not covered

This covers what your observability tool captures and ships to its own backend, not the text on its way to the model provider itself, and not a hand-written log line your own request handler writes straight to disk.

Leans on

6 steps

Where to go next

Redact PII from a message before it reaches a third-party LLM

want to stop the model provider itself from ever seeing the real values, a stricter goal than keeping them out of your own observability tool

Raw LLM logs become a second copy of your customer data

want to handle a hand-written log line your own code writes straight to disk, instead of a trace your observability tool auto-captures and ships

Checked 15 Aug 2026

Part of the LLMOps cookbook