Blur or black box? How to hide text in a screenshot safely

Use a solid, opaque box. Do not blur and do not pixelate: both only scramble the letters, and researchers have turned that scramble back into text. A filled rectangle, exported as a new flat image, leaves nothing underneath to recover.

Why is blur or pixelation weak for hiding text?

Pixelation cuts the image into small squares, each with the average colour of the pixels it covers. Blur mixes each pixel with its neighbours. Neither removes the letters. Both turn them into a new pattern that still depends on the text. And text is an easy target, for four reasons.

• The alphabet is small: a few dozen letters and signs, plus ten digits.
• Fonts are known. An attacker can install the font a website or an app uses.
• Blocks are independent. Bishop Fox explains that one pixel only affects its own block, so the text can mostly be guessed character by character.
• Guessing is cheap. A program pixelates each guess the same way and compares. A known format, such as a six-digit code, narrows the guesses.

Has anyone really recovered blurred or pixelated text?

Yes, in public. In 2016, Steven Hill, Zhimin Zhou, Lawrence Saul and Hovav Shacham published a study in the journal Proceedings on Privacy Enhancing Technologies. Using hidden Markov models, borrowed from speech recognition, they read pixelated and blurred text, including real-world examples. Their verdict: mosaicing (pixelation) and blurring are not viable approaches for text redaction.

Tools followed. In February 2022, Dan Petro of Bishop Fox released Unredacter. It solved a public challenge from Jumpsec Labs, a text pixelated in blocks of 5 by 5 pixels, and Jumpsec confirmed the answer.

When you need to redact text, use black bars covering the whole text. Never use anything else. No pixelization, no blurring, no fuzzing, no swirling. (Dan Petro, Bishop Fox, February 2022)

To be fair, these attacks need skill and the right font, and they often fail. But nobody knows which tools will exist in five years.

What is the safe way to hide text in a screenshot?

Cover the text with a filled shape in one solid colour. Black is usual, but any fully opaque colour works. The box replaces the pixels, so nothing is left to average or compare.

1. 1Draw a filled rectangle, not a pen stroke, at 100 percent opacity.
2. 2Make it wider than the text. A tight box reveals the length of the word.
3. 3Cover every copy: tabs, notifications, account names, other people's details.
4. 4Flatten: export a new PNG or JPEG, so the box and the picture become one layer of pixels.

Is cropping safe? What about thumbnails and metadata?

Cropping should be safe. In 2023 it was not, on two common tools. Google's Pixel bulletin of March 2023 lists CVE-2023-21036, a high-severity flaw in Markup, the Pixel screenshot editor. Kaspersky credits Simon Aarons and David Buchanan with the discovery, nicknamed aCropalypse. Buchanan explained that the cropped image was written over the old file without truncating it. On one of his own screenshots, the leftover data revealed his full postal address.

Microsoft's advisory CVE-2023-28303, published on 24 March 2023, covers Snip and Sketch on Windows 10 and Snipping Tool on Windows 11. Microsoft rates it low: the image must be saved, cropped, then saved to the same location. Its example: a bank statement with the account number cropped out. Versions 10.2008.3001.0 and 11.2302.20.0 fix it. According to Kaspersky, updates cannot fix images already shared.

Thumbnails are a second trap, mostly for JPEG photos, which can hold a small preview in their EXIF metadata. In a 2016 Kaspersky test, one photo editor kept the original preview after an edit. EXIF can also reveal the date, the camera and the GPS position, for example when you photograph a paper with your phone.

What is a safe routine for every screenshot?

1. 1Ask whether you need a screenshot at all. Sometimes one sentence is enough.
2. 2Capture only the window or area you need.
3. 3Keep the original in a private folder and work on a copy.
4. 4Hide each sensitive item with the four steps above.
5. 5Save to a new file with a neutral name, not invoice-jane-example.png. Never overwrite.
6. 6For a photo, remove the metadata. Kaspersky suggests an image viewer or a dedicated app.
7. 7Open the final file, zoom in and read it as a stranger would.

In the EU, this is also a legal matter. Under Article 4(1) of the GDPR, personal data is any information relating to an identified or identifiable natural person, such as the name of a client, Jane Example. Article 5(1)(c) limits data to what is necessary for the purpose. Hiding what the reader does not need applies this data minimisation rule.

A box does not make the whole image anonymous, though. A job title or a rare detail left visible can still point to one person. Recital 26 of the GDPR treats data as anonymous only when the person is not or no longer identifiable.

When does a dedicated tool help?

For one screenshot, your usual editor is enough if it draws opaque shapes and exports a flat image. For twenty screenshots, a scanned contract or a long PDF, finding every name, IBAN and email by eye is slow, and tired eyes miss things.

Detection tools help there. Desktop PDF editors offer redaction; check in the vendor's documentation what the export really removes. ONYRI Sanitize is one browser-based option: detection and masking run in the browser, with no upload. Its Black marker mode puts opaque bars on a flattened export. Its on-device OCR reads English, French and German, not handwriting. Detection is never exhaustive, so review each item before downloading.