machinewitness

Independent · European · without judgement

What websites tell machines, sealed day by day.

Whether an AI company may use content is set out in files nobody ever sees, and they can change overnight without a trace: robots.txt, ai.txt, TDM reservations. Since 22 July 2026, two independent witnesses have recorded every day what around 139,000 European domains publish in them, and every day is sealed and anchored beyond our reach.

If you ever have to show what your site told AI crawlers on a past day, the days are already on record here.

total observations sealed72,698,451
latest seal2026-11-17 (772,597)
anchoringOpenTimestamps · RFC 3161 · qualified eIDAS
active since2026-07-22

Check your domain: what is possible today The check runs on a hash your browser computes. No search engine, no directory of what we hold.

The evidence extract

The evidence extract: what your domain said on the day

Everything recorded on one domain, every day since it entered the archive, from both witnesses, as one document you can file. Any expert can recompute it without us.

Example: a reservation was already in place when the content was used. The extract shows every day it appeared in the file.

450 €per domain, everything recorded on it

How the archive is checked for a domain and what an extract contains, shown with an example case. Voice and opening motion: AI-generated.
Transcript

Websites change all the time. What was there yesterday, prices, terms, a statement, is often gone today.

MachineWitness records every day what around 128,000 domains in the EU serve: two independent witnesses in two countries, every day sealed.

What is stored is exactly what the website served, byte for byte, with the time of the fetch and a digital fingerprint.

If a dispute comes months later, the evidence extract takes those days out of the archive, as a document a court can check too.

An example: the firm Byrne & Walsh represents a client. It is about the price shown in August on a provider’s homepage. Today a different one is shown.

The domain does not have to belong to the firm. Anyone can order an extract, the other side too, on the same terms.

First the firm checks for free whether the archive has the domain. Here we show it with our own domain.

The check also says how: for most domains, the files for machines every day and the whole homepage once a week; for domains in daily observation, everything, every day.

The extract costs 450 euros plus VAT per domain and contains everything the archive holds on it. We do not ask what for.

Payment is by card. Delivery is normally within five working days; if a court deadline is running, within 48 to 72 hours.

The first page says openly how the domain was recorded and which days are included.

Then every day: every file as it was served, from both witnesses, with the daily seal and a qualified timestamp.

With the enclosed instructions, any expert can check it themselves, without having to ask us.

Evidence extract from MachineWitness. It shows what was there. What follows from it is for the parties to decide.

For proceedings: a technical procedure statement and written technical answers.

Publisher or media house? Your AI reservation in robots.txt can be shown for every past day the archive holds.

Why this matters in practice

Under EU copyright law, a text-and-data-mining opt-out only counts if it was machine-readable at the time (OLG Hamburg, 10 Dec 2025). The hard part is the date. An AI answer can show that content was read, never on which day. A screenshot or a server log shows one moment, and it comes from the party that benefits from it. The files themselves can be changed overnight without a trace.

What we prove, and what we do not. We prove that the reservation stood, on which days, exactly as it was served. That a particular piece of content went into a particular model is not something we can show; the claimant establishes that separately. We are one strong link in the chain of evidence, not the whole chain. The AI company and the court see the same record, on the same terms.

What we do not do

We do not judge, rank, score, accuse, or advise. We do not decide who is right. We record what was served and seal it, and nothing more.

A witness, not a detective. The value of the record is precisely that it takes no side.

How the proof holds

  1. Raw record: exactly the bytes the server sent, with headers and TLS fingerprint.
  2. Daily seal: each day as one Merkle root that anyone can recompute.
  3. External anchors: RFC 3161, OpenTimestamps (Bitcoin) and a qualified eIDAS time stamp.
  4. Second witness: since 4 August 2026 a second machine with another provider in another country.

How it works, step by step · Public log: every sealed day · What we store about a domain

Who is behind it

Martin Schenk

The archive is built and operated by Martin Schenk, engineer (Dipl.-Ing. (FH), automation engineering), who has lived and worked in Madrid for over thirty years. He develops and runs the entire system: crawler, sealing, anchoring, and the two witnesses. The technical work, and the documents this archive issues, are produced with the help of AI under human supervision.

Benjamin Schlunk

Benjamin Schlunk, Jurist (Rechtsreferendar, German legal trainee), accompanies the project in legal matters. He studied law in Göttingen with a focus on commercial law, and holds the First State Examination in law. His role is to review this archive's own legal position: its lawful basis, its deletion procedure, and how its records should be described. He does not advise users of this archive, and nothing this archive issues is legal advice or an opinion on anyone's case.

The archive is built so that you rely on its procedure, not on the people running it: a third party can check everything it does without involving us.

Disclosure, so neutrality can be judged openly: should the operator ever advise or represent a party that relies on a MachineWitness record, that involvement is disclosed. The archive takes no side in any dispute.

Also from this archive: daily observation of a domain from today on · a point-in-time record of named addresses · for individuals: AI training, yes or no.