Capture with consent
Recording starts only after explicit, document-level approval — on every surface. Off by default, revocable anytime.
Not a detector’s guess: a consented, sealed record of how work was written, so universities can verify authorship and administer AI-use policies, and anyone can check it.
The railway did more to reshape the nineteenth-century town than any act of parliament.
By 1851, more than half of the population lived in towns, a figure I compare with the parish records in the section that follows
Bibliography to follow…
Two products
Schools and universities
Writing provenance and evidence infrastructure for universities: verify authorship, administer AI-use policies and manage academic integrity risk with consented process evidence, and give students a fair way to show how they worked.
Teams and professionals
Verifiable authorship and compliance records an auditor can check, priced per seat. The writer sees and owns their full record and shares it with clients and regulators.
Verifiable process evidence for integrity reviews and appeals, with its limits stated and less reliance on detector scores.
A way to show their own work. Recording is off until they allow it for a document.
Verifiable authorship, and records the writer can see, share and an auditor can check.
A tamper-evident seal anyone can check, with a named person who attested.
The problem
Detectors read finished text and guess a probability it was machine-written. When the guess is wrong, an honest student carries the accusation, and the institution has to defend a probability in a hearing or an appeal. Rewriting a finished draft changes what the detector reads, so the guess does not get better under scrutiny. An integrity decision needs a record of how the work was written.
The authorship verification category includes output detectors that score finished text and writing-replay tools that record an edit log. Authmarks is a third approach: record the process as it happens, with consent, have a named person attest, and seal a record anyone can check.
Guessing after the fact
Statistical detectors are probabilistic: they produce a score, not a record anyone can check. A wrong score can lead to a false accusation, and detectors get less useful as writing tools get better.
The Authmarks approach
Record the writing process while the work is created, then seal it into a tamper-evident record anyone can check, with a named person attesting before sealing. A reviewer reads what happened; the decision stays with people and the institution’s own policy.
Academic integrity risk
An integrity matter has two parties, and both are exposed when the evidence is a guess. Authmarks gives the institution verifiable evidence, with its limits stated, for a review or an appeal, and gives the student a fair way to show how the work was written. The record states what was observed. People decide, under the institution’s own policy.
For the institution
Due process rests on evidence both sides can examine. A sealed record of the writing process is that evidence, and it does not depend on a detector’s guess.
For the student
A student who wrote their own work should not have to prove a negative. With a record, an honest writer can show a reviewer how the work was written.
Source lineage
A pasted paragraph rarely starts where it lands. It may have been copied from an article into a notes file on Monday and pasted into an essay on another computer on Thursday. Source lineage shows that path, one step at a time, back to the earliest place the copy was witnessed.
Illustration only, not real data.
The limits, stated
Patent-pending methods: the methods behind source lineage and the character-level lifecycle are covered by provisional patent applications, so we describe what they show, not how they work.
Workflow
From the writing to a record an instructor, an integrity panel or a client can check, without leaving the document you already write in.
Recording starts only after explicit, document-level approval — on every surface. Off by default, revocable anytime.
The record shows how the writing arrived while recording was on — typed, pasted, suggested, or untracked. It never reduces the document to a single score.
Authors or designated reviewers confirm accountability before high-trust outputs. A named person stays in the loop.
A sealed record anyone can check. The fuller writing record stays with people the owner authorises.
The platform
Authmarks witnesses writing wherever it happens. Every surface feeds the same sealed record — one record, one attestation, one public check.
Browser extension
One extension witnesses writing in both editors — the tools people already use, with nothing moved and nothing re-formatted.
Native add-ins
A native add-in for Word and an add-on for Google Docs will bring the record into the editor itself — no browser extension required. Coming soon.
Authmarks Write
A native desktop app and a web writing room with its own document drive, live collaboration, and the capture record built in from the first keystroke.
Phone app
A phone app for writing is coming soon.
Liveness signals
Typing-rhythm analysis and voice-matched dictation generate signals about whether physical input, not script injection, produced the text — signals for a reviewer, not a verdict.
Public verify
No account needed. Anyone can check a sealed record at authmarks.com/verify. The cryptographic check verifies our signature is intact — it does not depend on your account or ours.
Diligence scorecard
A sealed record should be able to answer these, in plain language.
The methods behind the record are patent-pending, which is why this page describes what the record answers — not how it is produced.
The record, up close
A person writes — in Docs, in Word, or in the Authmarks editor. A named person stands behind the record. Then it is sealed, so a third party can check that it still matches.
Recording is on only because the writer allowed it for that document.
The author, or a reviewer they choose, accepts responsibility for the record.
After sealing, someone can check that the record still matches what was sealed.
Illustration only.
Education
Institutions are switching detectors off after false positives and contested hearings — but they still have to show who did the work.
Australian Catholic University reported around 6,000 academic misconduct cases in 2024, with roughly 90% AI-related; cases relying on detector-only evidence were dismissed.
ABC News · 9 Oct 2025
Washington State University cancelled its AI detector in February 2026; about one-third of related hearings concluded “not responsible.”
WSU provost memo
Ninety-four percent of UK students report using AI in assessed work.
HEPI 2026
Curtin University disabled its AI detector from 1 January 2026.
Curtin University
For business
For organizations and the professionals inside them, anywhere a document’s authorship has to survive scrutiny. The writer sees their own full record, a named person stays accountable, and a sealed record can be shared with a client or a regulator who checks it themselves. Authmarks for business →
Enterprise & regulated industries
A record of human review for regulated content, of the kind auditors and Article 50 transparency duties ask about, captured as the work is done.
Legal & professional services
Provenance and reviewer accountability for contracts, filings, and client deliverables in high-scrutiny workflows.
Media & publishing
Show how content was produced and reviewed before publication — the human-oversight record regulators increasingly expect.
Consulting & agencies
Hand clients work with a verifiable record of how it was produced — a differentiator when anyone can generate a polished draft.
Protection
The methods behind the capture, the record, and the seal are patent-pending.
Status
The capture, source lineage, fusion and sealing methods are covered by provisional patent applications.
Access
A free plan is available — sign in with Google and start sealing your own work. Institutions and teams write to sales@authmarks.com.
In the browser
One extension witnesses both editors. A named person attests.
Beyond the browser
Our free desktop app witnesses typing and copy-paste on the computer, keeping key text only in programs where a document can be. Native add-ins are coming soon.
Human effort model
Authmarks classifies tracked writing by category and keeps untracked text explicit — so reviewers can assess process integrity without reducing the work to a simplistic “AI / not AI” verdict.
Written by a person during the session.
Brought in from outside the session.
Not written by a person at the keyboard. A label for review, not a verdict.
Inserted by software rather than typed.
Accepted from a suggestion while writing.
Spoken into the document; the recording is checked, against the writer’s own voice once they have set it up.
Grammarly or Gemini, when the change is witnessed as coming from the tool.
Written by another person recording the same document.
No recorded provenance — never silently attributed to a human.
How it’s verified
A record only helps if nobody has to take it on trust. Underneath every Authmarks record is a seal built on open standards, so an integrity panel, a client or a regulator can check it for themselves.
Integrity
The writing events are hash-chained as they arrive and the finished record is sealed with a C2PA signature. A change after sealing shows up as a failed check.
Authenticity
Each seal names the author who signed the declaration, chains later reviews to it, and carries an RFC 3161 time-stamp of when it was signed.
Chain of custody
The seal records which connections submitted the evidence, over what period, and any mismatch between the signed-in account and what a device reported.
Independent check
Anyone can check a seal at authmarks.com/verify with no account, or with standard C2PA tools against our published root certificate.
A valid seal shows the record has not changed since it was sealed. It does not decide what happened; that is for the people reviewing it. Read about the evidentiary standards →
Human review before seal · the heart of accountability
This is the core of what makes an Authmarks record trustworthy. Before a session can be sealed, a person reviews the work and accepts responsibility for it — either the author reviewing their own work, or someone they invite to review it. Authmarks records who that person is and that they attested, and the seal stays locked until they do. It’s how the record carries clear editorial responsibility, in line with EU AI Act-aligned principles of keeping a natural person in control wherever AI contributes.
Step 1
When the writing is complete, open Human review from the session before sealing. The seal is gated behind this step.
Step 2
Two paths, one principle. Review your own work and stand behind it, or send an invite so a second named person reviews it and takes responsibility. Either way, accountability has a name attached.
Step 3
The reviewer confirms they’ve reviewed the work and accept responsibility — accounting for any AI or automation involved — and can leave a note for context.
Step 4
Only once review is complete does the seal unlock — so the final, verifiable record always carries a human who stands behind it.
Security & privacy
To record how work was written, Authmarks has to process the writing it records. The point is not to hide it from us; it is to make the result something a reviewer can check, while collecting little and leaving the writer in control of recording.
Each sealed record can be checked by a third party. A change after sealing is visible as a failed check. The fuller writing record stays with people the owner authorises.
Recording is explicit and per-document. Nothing is recorded until you allow it for that document. You can turn recording off.
Unsealed session telemetry is short-lived by default. Sealed provenance may be retained for verifiability, legal, contractual, or anti-fraud reasons. In-region hosting and organisation-held keys are available as enterprise options.
The moment
Institutions are moving away from detector scores and toward evidence they can explain in a review or an appeal. For business, since 2 August 2026, AI-generated text published to inform the public must be disclosed, unless a person reviewed it and holds editorial responsibility. Authmarks’ named attestation is a record of that review. For generative systems already on the market before that date, Article 50(2) machine-readable marking has a grace period to 2 December 2026.
Platforms & partners
The sealed record lives with the document. From there, a partner can surface it in an e-signature workflow or a learning system.
LMS platforms such as Canvas connect today via LTI 1.3. REST API integrations are scoped with partners. The capture surfaces already span Google Docs, Word, the desktop app, and our own editor; deeper integrations follow the engagement.
Teachers start free with one class. Businesses get a 14-day trial of Pro or Team with no card. Institutions: one agreement covers every course.