We didn't start here.
But the problem found us.
/Draft exists because writers who do honest work deserve better than a probability score.
From photo authentication
to proof of authorship.
Signet (Signet Proof, Inc.) started as a photo authentication tool. Generative AI was and still is saturating every online space. I've worked in digital media for over 20 years and can no longer answer with any confidence when asked: “Is this realAI? The trend was clear. The trust relationship with digital content was already strained, but now it's completely broken. Default skepticism must become the only rational position.
Detection is a losing game, the models will always be a generation behind. There is no way to prove something wasn't created by AI, using AI. It will never work.
I was building a way to document proof of human creation in digital media, cryptographically protected from tampering, for situations where the stakes of getting it wrong mattered. When trust in what you are seeing is required. The technical model was straightforward: capture what happened at the moment it happened, how it happened, in a way that anyone could verify.
My partner works in higher education. They were seeing the failure mode firsthand. Students struggling with how to use AI, detection tools flagging the work they did as plagiarism. A colleague told a student to change the way they wrote because their natural voice 'sounded like AI phrasing.' Whether the student wrote every word didn't matter.There was no credible way to tell the difference.
The absurdity was playing out in real-time. The Gettysburg Address scored 97-100% AI-generated by ZeroGPT. At Adelphi University, a student falsely flagged by a detector won a lawsuit after their family spent six figures defending the human-written essay. Cases like these were piling up, students being penalized for work they actually did, on the basis of a probability score that couldn't distinguish their writing from machine writing.
The architecture I'd been developing applied directly. Not by trying to detect AI in finished work, but by giving the person who did the work a way to prove they did it. This was the same trust problem in a different domain, but in a situation the stakes mattered.
Instead of asking "does this look human?" — a question no tool can reliably answer — we ask "can you show you made it?" That question has a real, verifiable answer. The burden sits where it should: not on the institution making accusations, but on the creator presenting evidence.
TRUST, NOT SURVEILLANCE
The relationship matters.
The arms race between AI detection and AI generation has done irreparable damage to something more important than either tool: the trust between students and the people who teach them.
Suspicion by default.
The conversation that should be about writing becomes a conversation about who's lying. We start from a different direction. The student creates the record. The student decides when to submit it. The instructor reviews evidence the student volunteered. Not a probability score generated without consent. A visual representation of a non-linear process. Data surfaced to be something more useful than a guess.
We didn't build this to catch cheaters. We built it to protect the students who do the work.
Commitments.
Built into the architecture from day 1.
These constraints are architectural and enforced by the structure of the system, not by policy statements that could be quietly changed. They reflect the particular tensions that exist in the academic context.
Not an institutional surveillance tool
/Draft is installed by you, not your institution. Your professor doesn't get access to your google drive. Your writing and record isn't accessible to anyone unless you hand it to them directly. The certificate is yours, for your work, not attached to a course, not locked to a platform, not visible to anyone who didn't receive it from you.
Not sold over your head
We want institutional adoption, that's how /Draft reaches more students. But we won't get there by selling a monitoring system to schools first. Your work belongs to you. Always.
Not a detection tool
/Draft makes no claims about AI use. It does not detect AI-generated content, and we have deliberately built it not to. It records a writing process record and presents the data. Academic judgment remains with the instructor. That's the design.
Not a content scanner
Document text never leaves your browser. Signet records a mathematical fingerprint of your document at intervals — enough to prove it existed in a specific state, never enough to reveal what it said. This is enforced by the system's architecture, not by policy.
Not vendor-locked
The certificate uses open cryptographic standards. The timestamp is issued by independent third-party infrastructure, not Signet's servers. The verification page runs entirely in your browser — no account, no software, no upload — and the cryptographic record uses standards that exist independently of us.
No Ethical conflict
Signet can say — and mean it commercially, not just rhetorically — that it doesn’t matter whether a student used AI assistance in their writing. What matters is that the student engaged in a real human creative process. The certificate records that process. The student's choices about AI use are their own.
Want to try it or talk about it?
What is Signet?
Signet ( Signet Proof, Inc. ). is the company building provenance infrastructure for accountable digital creation.
/Draft is the first product, it focuses on student writing. Our other products in development extend the same trust model to video and creative work. The underlying infrastructure is built on open standards. (C2PA, CAWG), with an open-source verifier and binding commitments to keep verification working even if the company doesn't.