# Cloudflare DLP says goodbye to source code false positives!

Hey everyone, it's me, Shii-chan! Today I found a quiet but genuinely nice security update, so let me walk you through it.

## What was announced?

On the Cloudflare Changelog, they announced improvements to Data Loss Prevention (DLP)'s **source code detection**. Going forward, it focuses on identifying whole source code file uploads and downloads.

## The story so far

Until now, source code detection ran partial scans that only looked at part of a file. That meant code snippets embedded in chat messages, documentation, or code samples often got mistakenly flagged as source code too.

## What changes

The new approach only evaluates standalone source code files. Code embedded in chat messages or documents is no longer treated as source code, so you should see a lot fewer false positives. For security teams running DLP, that means less alert fatigue.

## Dive Deep

A few concrete details were shared too.

- Source code detection only evaluates files with **at least 500 characters**. Anything shorter is filtered out as noise
- Sensitivity can be tuned with **confidence levels** — a higher level cuts false positives, while a lower level catches more files at the cost of extra noise
- This applies to standalone source code files in **Gateway HTTP policies** only. It doesn't detect code embedded in other file types or payloads, like `.docx` files or chat messages, so keep that in mind

## Wrap-up

- DLP's source code detection moved from partial scans to **evaluating whole files**
- Code embedded in chat messages or documents is no longer mistakenly flagged
- Only standalone files of **500+ characters** are evaluated, and sensitivity is tunable via confidence levels
- Scope is limited to Gateway HTTP policies — embedded code in other file types isn't detected

This is a nice one for security folks running DLP in Cloudflare One!