Does AI leave a watermark?Sometimes, and marks can be lost; absence does not prove a file is human-made

TimeBinder adds a layer of protection against AI and human fraud and disputes.

The file stays on your device. Your browser makes a digital fingerprint of the file. TimeBinder records the fingerprint and the time, not the file.

Short answer

Sometimes. Some AI tools attach a mark, an invisible watermark such as SynthID, or Content Credentials. Many do not, and marks can be lost in edits or re-uploads. A missing mark does not prove a file is human-made. A TimeBind does not look for watermarks; it dates an exact file.

The different kinds of “watermark”

People use “AI watermark” for several different things. A visible logo or caption is the obvious kind, and it can be cropped or covered. SynthID is Google DeepMind’s invisible watermark for some generated images, audio and, in some products, text; it is read with a matching detector, not by eye. Content Credentials are not a watermark: they are a signed C2PA label. See what Content Credentials are.

For text, a watermark is usually a hidden statistical pattern in the words a model chooses, not hidden characters you can see or delete. As of the sources checked for what AI text watermarks mean for writers, Anthropic has said it marks text from supported Claude models, and OpenAI has said it will add a watermark to eligible ChatGPT and Codex text in the EU. Detection, in those public statements, is restricted to approved organisations, not a tool you can run yourself. Other products may add a mark, a label, both, or neither — and that changes.

The AI watermark and label decoder is a plain-language guide to what a label or detector claim might mean. It does not detect or remove watermarks.

What a missing mark means, and what a TimeBind shows

If you cannot find a watermark or a Content Credentials label, that does not prove the file is human-made. Many generators do not mark their output, and marks are often lost in edits and re-uploads. A detected mark can be a useful clue; it is not, on its own, a complete account of how the file was made.

A TimeBind does not look for watermarks. It records a SHA-256 fingerprint of the exact file you hold and a time, made in your browser, with the file not uploaded. It shows that file existed no later than the recorded time and still matches. If you write or shoot your own work, date your originals — including drafts from before any AI pass — so you have a record that does not depend on a watermark surviving.

Make a TimeBind: first one free →

First Pooled TimeBind free, no card, then from US$5. An ASAP TimeBind is US$15.

Frequently asked questions

What is SynthID?

SynthID is Google DeepMind’s invisible watermark for some generated media. It is designed to be read by a matching detector, not by looking at the file. It is not the same as Content Credentials.

Do ChatGPT or Claude watermark their text?

Public statements summarised on our AI text watermarks guide say Anthropic marks text from supported Claude models, and OpenAI has said it will add a watermark to eligible ChatGPT and Codex text in the EU. Detection is restricted. Check those companies’ current pages, because this changes.

If there is no watermark, is the file human-made?

No. Many AI tools do not attach a mark, and marks can be lost when a file is edited, compressed or re-uploaded. Absence is not proof.

Does a TimeBind detect or remove an AI watermark?

No. A TimeBind records a fingerprint and a time. It does not detect, apply or remove watermarks, and it does not show whether a file was generated.

Sources

Google DeepMind, SynthID.
TimeBinder, What AI text watermarks mean for writers (Anthropic and OpenAI public statements cited there).
C2PA / CAI, Content Credentials.

Related guides and tools

Your file’s fingerprint, anchored to a global network secured by real electricity and cryptography.

Editing a file—including with AI—changes its fingerprint, not its existing network record.

How this works

Your file’s SHA-256 fingerprint, a short code worked out from its exact contents, is made in your browser, and the file itself stays on your device. A cryptographic commitment to that fingerprint, which may be combined with others, is then anchored in a block on the Bitcoin blockchain. Bitcoin is secured by proof of work: computers around the world spend real electricity to add each block, and changing an earlier block would mean redoing that work faster than the rest of the network. Until that block is confirmed, your proof is pending. Once the Bitcoin anchoring is confirmed, you can verify the proof against that recorded commitment. The confirmed record can then be checked; Pooled TimeBinds are checked through TimeBinder, while ASAP TimeBinds can also be checked independently. If you check the file later and its fingerprint still matches, that shows the file existed no later than the recorded time. The record doesn’t show who made the file, whether AI was used, or that its contents are true.

Your filestays on your device
FingerprintSHA-256
Network timestampon Bitcoin

Record a fingerprint of the file you rely on

Add a layer of protection against AI and human fraud and disputes: a dated, checkable record of the exact file you hold.

Open the app and sign in with email or Google.
Add your files. They stay on your computer.
Seal them and download your Proof of Time certificates.

Open TimeBinder →

First Pooled TimeBind free, no card, then from US$5. See pricing.