Authorship Certificate
Authorship Certificate: AI-detection defense for ESL writers.
The Diglot Authorship Certificate is an AI-detection defense tool for writers of English as a second language: a cryptographically signed, publicly verifiable record of how your document was written, designed for false AI-accusation defense in admissions, journals, and portfolios.
Independent research shows AI detectors are wrong about non-native English writers more often than they are right. Diglot logs your writing process as a cryptographically signed event chain, so when a teacher, editor, or admissions reviewer questions whether your work is AI-generated, you share a certificate URL — they verify the signature in their browser, no account needed.
*No credit card required
Authorship Certificate
№ AC-2026-0514-1840
This certifies that the document referenced below was edited inside Diglot during a recorded period, with an edit pattern consistent with original human drafting.
Signature verified
Simplified preview — real certificates include the full payload, surface mix, and tamper-proof disclaimer.
- Cryptographically signed — anyone can verify, no Diglot account required
- Records edits, pastes, and AI-assists as a tamper-evident chain
- Designed for false AI-accusation defense (admissions, journals, portfolios)
How the certificate works
Short answer Diglot records your edits as an append-only chain while you write, signs a certificate with an ed25519 private key when you request one, and gives you a public URL that anyone can open to check the signature. Three things happen, in order, and you don't have to do anything special during writing.
- 1
You write as usual
Diglot logs every edit, paste, and AI-assist as an append-only event chain. The log records what changed and when — not your full document content. Nothing extra to do.
- 2
You request a certificate
When you submit work that might be questioned, you generate a certificate from your document. Diglot signs it with an ed25519 private key and assigns a tier — Green, Yellow, or Red — based on the edit pattern.
- 3
You share the verification URL
The certificate is a public URL with a unique ID. Anyone with the link — your teacher, editor, or admissions officer — can open it and see a cryptographic check against our public key. No login. No Diglot account on their side.
What's inside a certificate
Short answer A Diglot certificate carries four things: the verdict tier, the writing period and word count, the Diglot surfaces that touched the document, and an ed25519 signature with the URL of the public key used to check it. It is a structured payload, not a vague "AI-or-not" label — everything is in the signed body, with no hidden state.
Verdict tier
Green / Yellow / Red — based on documented thresholds for paste size, AI-assist ratio, edit cadence. Same tier rules every time; not a black box.
Period & word count
When the writing started, when it ended, and how many words appeared in the final document. The cert covers a specific time window.
Surfaces
Which Diglot surfaces touched the document — web editor, browser extension, Google Docs add-on, mobile, or desktop overlay. The chain unifies cross-device editing.
Signature & public key URL
An ed25519 signature over the entire payload, plus a pointer to /.well-known/diglot-pubkey.json on diglot.ai where the public key is published. Verifiers fetch the key and check the signature in the browser.
How verification works (third-party view)
Short answer Whoever receives a Diglot certificate opens the URL in any browser — no install, no login, no Diglot account — and sees the verdict tier, period, word count, and edit-pattern summary on one screen, plus a signature check that reads VERIFIED, INVALID, or OFFLINE.
When you receive a Diglot certificate URL — typically a link like https://app.diglot.ai/verify/<id> — here's what happens in your browser.
Open the URL
Same as any web page — no install, no login, no Diglot account.
See the verdict
Verdict tier, period, word count, and edit-pattern summary on one screen.
Signature check
Runs inline in your browser — verdict reads VERIFIED, INVALID, or OFFLINE.
Download PDF
Static copy if a reviewer wants the certificate attached to a record.
Read disclaimer
The certificate explains exactly what the verdict does — and does not — claim.
Why this exists
Short answer Seven commercial AI detectors misclassified 61.22% of ESL essays as AI-generated in a 2023 Stanford study; the University of Waterloo discontinued Turnitin's AI detector in September 2025; and the New York Supreme Court ordered an AI-cheating record expunged in January 2026. The Authorship Certificate is the response.
| Evidence | What it found | Source |
|---|---|---|
| Research · Stanford 2023 | Seven commercial AI detectors misclassified 61.22% of ESL essays as AI-generated in a 2023 Stanford study (Liang et al., Patterns, Cell Press), while native-speaker essays were flagged at roughly 5% — a 12× false-positive gap. | Liang et al., Patterns (Cell Press), 2023 |
| University policy · 2023–2026 | The University of Waterloo discontinued Turnitin's AI detector in September 2025, citing «unreliability and bias toward students whose first language is not English». Vanderbilt followed; MIT published guidance titled «AI Detectors Don't Work. Here's What to Do Instead». | University of Waterloo, September 2025 |
| Court precedent · Jan 2026 | The New York Supreme Court in Matter of Newby v. Adelphi University ordered an AI-cheating record expunged in January 2026, calling the university's decision «without valid basis and devoid of reason». A federal complaint against Palo Alto USD (docket 5:25-cv-04202, N.D. Cal.) over a Turnitin flag on a high-school essay is now in discovery. | New York Supreme Court, January 2026 |
| Vendor admission · Feb 2026 | Turnitin's own documentation warns of a ±15-percentage-point uncertainty band and states the AI indicator «should not be the sole basis for punitive action». In February 2026 the company publicly pivoted «from detection to transparency». | Turnitin's own documentation and February 2026 public pivot |
Deep dive with full citations: AI Detection Lawsuits 2026 — What ESL Writers Need to Know
Last updated
When you need a certificate
Short answer Use a certificate for high-stakes writing where a false AI accusation has real consequences — admission essays, journal submissions, portfolio pieces, and coursework. Not for routine drafts.
Admission essays
A graduate program flags your writing as AI. You email the verification URL with your application response.
Journal submissions
A reviewer asks for proof of authorship for a manuscript revision. You attach the certificate alongside the response letter.
Portfolio pieces
A hiring manager doubts a writing sample. You share the certificate before the interview.
Coursework
A teacher uses an AI detector and you scored a false positive. The certificate is the evidence-based counter to the detector.
Authorship Certificate FAQ
Short answer No Diglot account is needed to verify a certificate, the edit chain records edit operations rather than the full text of every keystroke, and the verdict tier comes from documented thresholds included on the certificate itself.
Is the Diglot Authorship Certificate an AI-detection defense tool for ESL writers?
Yes — that is what it is designed for. Independent research shows AI detectors are wrong about non-native English writers more often than they are right: seven commercial detectors misclassified 61.22% of ESL essays as AI-generated in a 2023 Stanford study. The Certificate is built for false AI-accusation defense in admission essays, journal submissions, portfolio pieces, and coursework. Instead of arguing with a detector score, you share a cryptographically signed record of how the document was actually written — and anyone can verify it without a Diglot account.
I am worried Turnitin will flag my essay as AI. What does a certificate do about it?
The certificate is the evidence-based counter to the detector, not a way to change its score. Diglot logs every edit, paste, and AI-assist as an append-only chain while you write, so the proof exists before the accusation — which is worth more than any counter-score produced afterwards. If a teacher runs an AI detector and you score a false positive, you send the verification URL: the reviewer opens it in any browser, with no login and no Diglot account, and sees the verdict tier, the writing period, the word count, and a cryptographic signature check.
Can an Authorship Certificate prevent a false-positive AI flag?
No, and it does not claim to. The false positive is produced by the detector; the certificate sits on the other side of it, as the record a reviewer reads after a flag in admissions, journals, portfolios, or coursework. What it prevents is having nothing to show. Because Diglot records the edit chain automatically while you write, the proof exists before the accusation instead of being assembled under pressure after one. The verdict supports your defense — it is not a verdict on guilt or innocence.
Do I need a Diglot account to verify someone else's certificate?
No. Verification is fully public. You open the certificate URL, see the payload and the signature status, and read the verdict — all without logging in. This is the core design property of the certificate: third parties shouldn't have to trust Diglot to act as a notary; they verify cryptographically.
What does the verdict tier actually mean?
Green, Yellow, Red are summaries based on documented thresholds — paste size relative to typed content, AI-assist density, edit cadence, and other signals. The thresholds are not opaque: the certificate itself includes the rules. A Green certificate is consistent with original drafting; Yellow flags significant AI involvement; Red flags patterns associated with generation. The verdict supports your defense — it is not a verdict on guilt or innocence.
What data does the edit chain record?
The chain stores edit operations (insert / delete / paste / accept-AI-suggestion) with timestamps and approximate sizes. It does NOT store the full text of every keystroke. The certificate aggregates the chain into a summary — the underlying chain stays inside Diglot unless you choose to expose it. See our Privacy Policy for the full retention rules.
Can someone forge a certificate to look like mine?
No. The ed25519 signature is over the entire certificate payload, signed with a private key that never leaves Diglot. The verification page fetches the public key from diglot.ai/.well-known/diglot-pubkey.json. A forged certificate would either fail the signature check (status: INVALID) or point at an attacker-controlled key URL (the verify page falls back to the canonical public-key location and would still mark it INVALID).
Does the certificate prove I wrote the document myself?
It proves the document was edited inside Diglot during a specific time window, with a specific pattern of typing, pasting, and AI assistance, on specific surfaces. That pattern is consistent with human drafting in the Green tier and inconsistent in the Red tier. The certificate is strong evidence; it is not a courtroom proof of authorship — and we say this in plain language on the certificate itself.
I do not have a certificate URL — how do I get one?
Write your document in Diglot, then request a certificate from the editor. Free-tier accounts can preview the certificate; full ed25519-signed certificates are part of the paid plans. See Pricing for plan details.
Latest research and case studies
Short answer The evidence base is public and named: the Stanford study on detector bias, the January 2026 New York ruling calling a university decision «devoid of reason», and the universities that disabled Turnitin AI detection. Every link below is verified-source coverage of detector bias, court rulings, and the practical defense playbook for non-native English writers.
- Litigation update · May 2026 AI Detection Lawsuits 2026: What ESL Writers Need to Know A federal complaint in California, a New York state ruling calling a university decision «devoid of reason», and prestigious universities quietly disabling Turnitin AI detection — the verified state of play.
- Research · Detector bias Why AI Detectors Misread Non-Native English The Stanford study showing seven commercial detectors flag 61% of ESL essays as AI even when fully human-written — and the four-step originality defense for writers who get falsely accused.
- Hub · If it already happened Flagged as AI? Start here The complete playbook: five steps to respond, the research compendium to cite in an appeal, per-detector guides — and why proof that exists before the accusation beats any counter-score.
- The science · Stylometry Your writing fingerprint: how authorship attribution works From the Federalist Papers to forensic linguistics: why style identifies authors, why detectors are crude population-level stylometry — and why process-based proof beats style-based guessing.
- Comparison · DIY vs signed proof Draftback and stronger alternatives for proving authorship What Google Docs replay tools do well, where they fall short as evidence, and what a cryptographically signed, append-only writing record adds.
- Practice · Before the accusation The authorship checklist What to keep, and from when, so the record exists before anyone asks for it — the habits that make a Certificate worth issuing.
- Browse all Authorship Certificate research category → Every article on detector bias, court precedent, university policy, and cryptographic authorship — updated as the litigation landscape develops.
Start writing with verifiable provenance
Free tier records the edit chain. Paid plans unlock full certificate generation with public verification URLs.
*No credit card required
