---
type: "post"
id: "the-open-source-unredact-tools-project-scans-files-5ad18f"
slug: "the-open-source-unredact-tools-project-scans-files-5ad18f"
title: "The open-source unredact.tools project scans files for hidden information and can rebuild sanitized copies"
description: "The open-source unredact.tools project scans files for hidden information and can rebuild sanitized copies, then scan those copies again."
retrieval_nugget: "The open-source unredact.tools project scans files for hidden information and can rebuild sanitized copies, then scan those copies again. For teams sending documents and media to AI services, it offers a practical pre-upload check: inspect what a file reveals, remove selected risks, and verify the result rather than trusting a visual edit alone."
published_at: "2026-11-16"
updated_at: "2026-11-16"
record_date: "2026-11-16"
date_kind: "published_at"
topics: ["anthropic"]
entities: ["unredact.tools","NPCmillionaire","Ollama","Anthropic"]
related_patterns: ["agent-security-moves-to-runtime-boundaries"]
novelty_label: "new_fact"
what_changed: "Unredact.tools introduces a file-analysis tool for finding and removing hidden metadata and content, including GPS coordinates, author names, deleted text and watermarks."
prior_refs: []
source_url: "https://github.com/NPCmillionaire/unredact.tools"
source_urls: ["https://github.com/NPCmillionaire/unredact.tools"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:the-open-source-unredact-tools-project-scans-files-5ad18f"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/the-open-source-unredact-tools-project-scans-files-5ad18f/","markdown":"https://newruntime.com/posts/the-open-source-unredact-tools-project-scans-files-5ad18f.md","json":"https://newruntime.com/posts/the-open-source-unredact-tools-project-scans-files-5ad18f.json"}
markdown_url: "https://newruntime.com/posts/the-open-source-unredact-tools-project-scans-files-5ad18f.md"
---

# The open-source unredact.tools project scans files for hidden information and can rebuild sanitized copies

## Retrieval answer

The open-source unredact.tools project scans files for hidden information and can rebuild sanitized copies, then scan those copies again. For teams sending documents and media to AI services, it offers a practical pre-upload check: inspect what a file reveals, remove selected risks, and verify the result rather than trusting a visual edit alone.

## Summary
The open-source unredact.tools project scans files for hidden information and can rebuild sanitized copies, then scan those copies again. For teams sending documents and media to AI services, it offers a practical pre-upload check: inspect what a file reveals, remove selected risks, and verify the result rather than trusting a visual edit alone.

## What happened
The unredact.tools GitHub repository describes a Python command-line utility for examining and sanitizing files. Its main commands are `scan` and `sanitize`. The scan command reports findings in plain language, grouped into categories such as location, identity, device, time, software, organization, hidden content, and tracking. A repository example shows a photo scan identifying GPS coordinates, a camera serial number, and a hidden thumbnail. Another example reports 39 leaks in a document before sanitization and zero afterward; those are examples in the project documentation, not an independent evaluation.

The project says it supports images, PDF, Office files, audio and video, and text. Scans can process multiple files, emit JSON, extract recovered artifacts, include lower-severity findings, or run in a faster metadata-only mode. Extraction can write recovered material such as embedded thumbnails, earlier PDF revisions, attachments, and embedded documents to an output directory. These capabilities make the tool more than a metadata viewer: it is intended to surface content that may not appear in the file’s visible presentation.

`sanitize` creates a clean copy rather than modifying the original, then automatically scans the output and lists anything that remains. The documented options include keeping selectable text in PDFs, blurring faces, redacting text or codes in images, redacting license plates, and selecting an image region. If a barcode remains, the example output identifies it and suggests rerunning with relevant redaction flags. Printer tracking dots are described as being removed automatically. The repository also documents a camera sensor-fingerprint feature and a sun-and-shadow command for estimating photo locations; these go beyond routine file cleanup and are separate investigative functions.

Installation is documented for macOS and Linux through a shell installer, or from source. The installer finds Python 3.10 or later and can ask before installing Tesseract and FFmpeg; it uses a virtual environment under the user’s local data directory and links commands into `~/.local/bin`. Optional extras add support such as HEIC files, visual checks, or AI scene review. The AI scene option uses Ollama locally by default. The README says an Anthropic API option uploads an image only after re-encoding it with metadata stripped, and warns the user each time.

## What the archive adds
We have not covered this before; it opens a new thread.

## Trend signal
This does not yet fit a pattern New Runtime tracks.

## Why it matters
AI workflows often move files across boundaries: from a shared drive into a model interface, from a customer conversation into a support tool, or from a local workstation into a vendor API. A document can carry context that its sender did not intend to disclose, including author identities, comments, previous revisions, or embedded objects. A photo can expose location or device details. Treating a file as safe because its visible page looks clean leaves these channels unexamined.

The useful design choice here is the scan-sanitize-rescan loop. It separates discovery from removal and makes the output a new artifact that can be checked. That is more actionable than a generic promise to strip metadata: a remaining leak is surfaced with a hint, allowing an operator to adjust the command. It also leaves the original intact, which is important for auditability and recovery. Teams can place such a check before uploads of sensitive material, but must still define which findings matter for their data and review the cleaned file’s visible content.

## Limits and uncertainty
The repository’s examples demonstrate intended behavior, not measured coverage across real-world documents or an independent security assessment. Detection and cleanup results will vary with file format, the way a document was produced, and the specific hidden material involved. A successful rescan is useful evidence about what this tool found, but it is not proof that a cleaned file contains no disclosable information or that its visible content is appropriate to share. Sanitization can also alter a file or remove information a team needs. The Anthropic API scene-review option creates a different privacy boundary from local scanning, even after metadata stripping. Teams should test representative files and apply their own retention, upload, and approval controls before relying on the utility.

## Skill to build now
Choose a small set of representative files that your team routinely uploads to AI tools: one PDF with comments or revisions, one image with location metadata, and one exported Office document. Run `scan` first and review the findings; then sanitize copies, keeping the originals untouched, and run a second scan. Record which findings matter for your policy and whether cleanup changed needed content. Use the results to decide where a file check belongs in the upload workflow, rather than treating a single clean scan as blanket approval.

## Reader takeaway
For sensitive files, visual inspection is not enough. unredact.tools makes a repeatable local workflow possible: scan for hidden information, create a cleaned copy, and scan it again. The practical gain is a more explicit checkpoint before a file enters an AI service; the remaining responsibility is to validate coverage and decide what your team is willing to share.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Anthropic](https://newruntime.com/topics/anthropic.md): Explore the anthropic topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)
