Patent-pending · Fraunhofer ISST

Prove every word is

Innamark places an invisible mark inside ordinary text. Readers see nothing different. You can still prove where it came from — whether it was written by AI, where it ended up, and who it was given to.

A spin-out research initiative from Fraunhofer ISST
00 The problem

Text travels. Credit doesn't.

A document shared once is shared forever. Your descriptions and articles show up on sites you never licensed. Nobody can say for certain whether a paragraph was written by a person or a machine. Photos and video have carried invisible marks for years. Text never had a way to prove where it came from.

0%
visible change to the reader
1mark
per recipient or channel
1paragraph
is enough to check
01 How it works

Four steps. Nothing changes for the reader.

  1. 01

    Start with your text

    Any finished text works as the carrier: a destination description, an article, a contract, an email. Nothing about it needs to be prepared or rewritten first.

  2. 02

    Add your watermark

    You choose what the mark stands for. One mark for a whole catalogue to follow it across platforms, or a separate mark per partner or per recipient when you need to tell copies apart.

  3. 03

    Distribute

    Publish or send it the way you always do — email, your content system, a shared drive, a partner feed. Every copy carries its mark wherever it travels.

  4. 04

    Verify

    Found your text somewhere you did not expect? Check it. Either your watermark is in there or it is not — a straight answer, not a probability.

Note How the mark is placed is protected under a pending patent. Licensing partners receive the full technical details under NDA.
02 Use cases

Three questions Innamark can answer.

Is this AI or not?

Traceability of AI content

Mark text at the moment it is generated, so it can always be recognised later as machine-written. The mark can also carry context — which system produced it, when, and under which policy — giving you a clear answer instead of a guess.

  • Tells AI-written text from human writing
  • Can carry model, date and policy details
  • Supports transparency and disclosure rules
Where was my content used?

Traceability of content usage

Central content providers — a regional tourism board, for example — supply descriptions and articles to dozens of guides, portals and apps. Mark each delivery and you can see where your content actually appears, which partners honour their licence, and where it is being reused without permission.

  • A separate mark for every partner
  • Shows genuine reach across channels
  • Surfaces unlicensed reuse
Who leaked this?

Traceability of leaks

Give every recipient of a sensitive document or email their own invisible copy. If it turns up where it shouldn't — with a journalist, a competitor, on social media — you can show which copy it was and who received it.

  • Individual copies for each recipient
  • Works on documents and email alike
  • Evidence you can act on
03 Research backing

Built on twenty years of applied research.

Innamark originates from one of Europe's largest applied-research institutions. The underlying technique was developed inside Fraunhofer ISST's data sovereignty group and is currently the subject of a pending patent application.

Origin
Fraunhofer Institute for Software
and Systems Engineering (ISST)
Status
Patent pending · DE/EP filing
Domain
Invisible watermarking for text
and data sovereignty
Licensing
Available to qualified partners
under commercial agreement

“The future of digital trust is provenance — not detection. Watermarking shifts the burden of proof back to where it belongs: with the source.”

— Innamark research team, Fraunhofer ISST
04 Frequently asked

Common questions.

Is the watermark really invisible to readers?
Yes. The marked text reads exactly like the original — same meaning, same tone, same length. Nobody reading it will notice anything, and there is nothing visible to spot or strip out. Only you, with your key, can check it.
What happens if someone paraphrases or translates the text?
The mark is built to survive normal handling: light editing, copy-paste, machine translation and rewording. The more of the original text remains, the more certain the result — a paragraph or two is usually enough to get a clear answer.
Can I try it on my own text right now?
Not publicly. The technique is patent-pending, so access runs through commercial licensing. Once we have spoken and an NDA is in place, you get a private environment to test with your own content, plus support from our team.
How does Innamark differ from existing AI watermarking schemes?
Most other approaches only work inside one specific AI system, and only for text that system produced. Innamark works on finished text, whoever wrote it — a tourism description, a press release, a board memo, or output from an AI tool you don't control. One approach covers all of it.
Who owns the IP, and what does licensing look like?
The invention belongs to Fraunhofer-Gesellschaft and is the subject of a pending patent application. Licensing is handled directly by the Innamark team at ISST, and is shaped around what you need — a pilot, a full rollout, or building it into your own product.
Does Innamark see or store my text?
No. Innamark runs inside your own environment, and the key that checks your text stays with you. Your content never has to leave your systems. Pilots can be run separately under standard data-protection terms.
05Get access

Request a demo.

Tell us what you want to protect or trace. We will get back to you as soon as possible with next steps and an invitation to try it.

Write to us

Helpful to include: your organisation, which of the three use cases fits, and roughly how much text is involved.

ResponseAs soon as possible
LocationDortmund · Online
AccessLicensing partners, under NDA