Who made it, how it’s governed, and exactly where it stands — stated the way the standard itself demands: plainly, with the limits visible.
Earned Trust was created by William Stafford, an independent researcher who spent his career as a nurse. He is dyslexic; AI tools gave him, late in life, the means to build ideas he had carried for decades — and that raised the problem this standard exists to solve: how could any reader tell what was his, what the AI did, and whether he had checked it? He built the answer he needed, used it on his own work first, and published it so others could use it too. His catalog at lostinnovators.org is the standard’s proving ground.
Earned Trust is a disclosure standard and a free toolset: a way to record which AI services contributed to a work, what each did, what the human directed and checked, and the evidence retained behind that account. It is self-declared, evidence-backed, and reader-checkable — self-declared meaning no outside body certifies it, never meaning self-invented. The author chooses to work under the Earned Trust protocol; the protocol keeps the record as the work happens and assigns the AIast disclosure from that record — the services, the roles, and the count come from the sessions themselves, not from memory afterward. The author approves the disclosure, answers for it, and can seal the finished work in a permanent archive with a DOI. It is not certification, accreditation, peer review, institutional endorsement, or a quality score. The records are what a reader weighs.
Today the standard has one maintainer: its author. That is stated rather than dressed up. How changes work:
Every change to the standard arrives as a new, dated, deposited version with its own DOI under the concept DOI 10.5281/zenodo.20719927. Nothing is ever silently edited; every prior version remains published beneath the newest. Corrections stand beside the originals, dated. The standard’s own working record — every session, ruling, and adversarial review behind it — is kept under the same rules it prescribes, sealed and producible.
Outside review is how the standard improves. Its versions to date were shaped by dozens of adversarial review rounds across multiple AI services, each finding reproduced before acceptance and each ruling logged — declines included, with reasons. Human reviewers are welcome and wanted: write to William@earnedtrust.org. If adoption grows, governance will need to grow with it; that evolution will happen in public, by new versions, like everything else here.
| Version | Date | What changed |
|---|---|---|
| v1.6 | 2026-06-16 | First public deposit (two same-day corrections, v1.6.1–v1.6.2; v1.6.3 followed 2026-07-26 with landscape additions) |
| v1.7 | 2026-07-29 | Built from adversarial review findings |
| v1.8 | 2026-07-30 | Evidence package, the redaction rule, conformance section, the trust distinction |
| v1.9 | 2026-07-31 | ETR edition: record page, seal, and QR carrying the work record |
| v1.10 | 2026-08-05 | Clean-signature edition + the sealed working record deposited with the standard itself |
| v1.11 | 2026-08-06 | The transparency dilemma confronted at full strength; the retention-discharge rule |
| v1.12 | 2026-08-07 | Evidence access clarified: the public manifest vs the on-request record |
| v1.13 | 2026-08-14 | The consolidation edition: one placement rule (author’s choice of three homes), the production middle path, the newest disclosure-trust evidence engaged, Codex as sixth disclosed entry — every change adjudicated across four pre-deposit review rounds, on the record |
The Field Kit runs its own line, v0.9.0 through v0.9.19 (current), with a changelog inside the kit; every prior kit version stays published on the kit page. Current versions are machine-readable at version.json.
Founder’s own use: five deposited works produced under the standard, each with its record, seal, and vaulted evidence. Independent users: one documented field record to date, produced by an unaffiliated author using the kit. Institutional pilots: none yet. Formal adopters: none yet.
We will not make a handful of users look like a movement. When those numbers change, this page will change, by dated edit. If you adopt the standard — as an author, a publisher, a department — tell us and be counted: William@earnedtrust.org.
The standard’s text is licensed CC BY 4.0; the tools are AGPL-3.0. Anyone may adopt the protocol and carry the AIast mark without permission or payment. The mark is not hand-picked: worked properly, the protocol assigns it from the record you keep — the minimum is the standard’s threshold: a record kept as you work, the disclosure drawn from it, and evidence retained. The LI- prefix is optional and self-applied. The mark carries no validity by itself — a dishonest mark costs the reader nothing, because the work and the producible records are the only things that ever mattered. Corrections to your own records follow the house rule: new dated versions, originals preserved. Conformance is self-declared, as the standard’s conformance section defines; there is no certifying body, and this page will say so for as long as that is true.
No accounts, anywhere. The Reader and the Field Kit process files locally on your device; nothing you drop into them is uploaded to us or anyone. Visit counting: this site uses GoatCounter, a cookie-free counter — no cookies, no cross-site tracking, no advertising identifiers, aggregate page counts only, which is why there is no cookie banner. The newsletter (when its signup is live) is consent only: you give an email address on purpose, it is stored with our newsletter provider (Kit), it receives occasional news and never your records or files, and unsubscribing or asking William@earnedtrust.org for deletion works the way you’d hope. Security contact and policy: security.txt.
Ten common questions, answered short: the FAQ.