EU AI Act · Article 50 · deployer duties apply from 2 Aug 2026

The compliance register for AI content.

From August 2026, the EU AI Act makes businesses disclose the AI content they publish. It does not require you to keep a record of that — but if an authority or a business customer asks whether you labelled what you had to, a record is how you answer. You declare once what’s AI. RegisterWell decides where a label is required, applies it, checks a visitor can actually see it, and writes every decision to an audit-ready register.

Heard the AI Act was delayed? Part of it was. On 24 July 2026 the EU published the Digital Omnibus, moving the high-risk rules to December 2027 and August 2028. Article 50 was not moved. It still applies from 2 August 2026.

What Article 50 actually asks →
your register, writing itself
IMG_7F2A · gen-img-v3
unlabeled
IMG_7F2A · gen-img-v3
● AI-generated · logged ✓
compliance registerrule-pack 2026.06
14:02:07Z · TXT_98D0 · editorial carve-out recorded · reason logged ✓
14:02:19Z · IMG_5C11 · platform stripped metadata · credential recovered · logged ✓
14:02:31Z · IMG_7F2A · label applied · gen-img-v3 · logged ✓
Why now

The rules are here. A spreadsheet won't survive them.

Article 50 splits the job in two: AI providers must mark content invisibly — and businesses that publish it must disclose it visibly and be able to answer for it when asked. Three facts to know:

3% of turnover

Transparency fines are capped at €15M or 3% of worldwide turnover. For SMEs and small mid-caps — which is most of the shops we work with — it is whichever is lower, so in practice the 3%.

Art. 99(6) & 99(6a), as amended by Reg (EU) 2026/1744
Conditional

The visible label isn't always required — deepfakes and public-interest text need it, human editorial review can lift it. Nobody can hand you a yes or no. The judgment, and the record of having made it, is the whole job.

Art. 50(1)–(4)
Aug 2026

Deployer disclosure duties land first — 2 August 2026. Provider machine-marking follows on 2 December 2026 for existing systems.

AI Act application timeline
How it works

Four steps. Zero new headcount.

01

Connect

On Shopify, install the app once — no theme surgery. Anywhere else, prove the domain is yours and we work from your public pages. Either way it is one step, once.

02

Know what's AI

Declare the AI tools and collections you use once — every future publish is then covered automatically. Where provider marks (C2PA) survive, we read those too. But we tested what actually happens: e-commerce platforms strip that metadata on upload. The mark your AI tool added is gone before a customer ever sees the page. That's where RegisterWell comes in — the decision is recorded the moment you publish, and a durable mark goes into the pixels, where stripping can't reach it. Honest by design: RegisterWell records and proves your disclosure decisions — it never claims to detect AI.

03

Label

The visible EU disclosure is applied where the law requires it — and skipped, with the reason recorded, where the editorial carve-out applies. Durable machine-readable marks stay intact underneath.

04

Prove

Every decision is written to the register automatically — asset, AI tool, timestamp, label, rule version. When a regulator or enterprise client asks, one click produces the evidence pack.

← the register is the product
Install once. It keeps recording.

On Shopify the app watches every publish: each asset is checked, the decision made, the visible label applied, the register written — zero manual steps, zero new headcount. Elsewhere we re-scan your public pages on a schedule and record what changed. Being designed hands-on with our founding partners.

First integration · Shopify · in design with founding partners
What you get

The decision, the label, and the proof you made it.

The audit-ready register

Asset, tool, timestamp, decision, rule version — written automatically, content-blind by design. One click produces the evidence pack a regulator or enterprise client accepts.

The right label, or a recorded reason

Visible "AI-generated" disclosure where Article 50 requires it — deepfakes, public-interest text, chatbot notices — and a recorded reason where the carve-out applies.

Rules that keep pace

Obligations live in versioned rule-packs. When EU guidance moves, your labeling and evidence update with it — no rebuild, no re-training your team.

Marks that survive

Underneath the register: C2PA credentials plus a robust invisible watermark, so provenance stays recoverable even after platforms strip the metadata.

Under the hood

The infrastructure is real. Break it yourself.

The register is only as strong as the marks beneath it — so we tested them where it counts. We uploaded a marked image to a live Shopify store: the C2PA credential was stripped by the CDN, and the invisible watermark survived intact, with full provenance recovered from it. That is the whole problem in one experiment — metadata does not survive commerce, and the moment it disappears, so does your ability to prove anything. That is where RegisterWell comes in. Try it yourself below: mark an image, push it through a platform pipeline, watch the credential die and the proof come back.

Instagram WhatsApp X Facebook LinkedIn Screenshot

Platform pipelines simulated on our robustness harness. First mark after a restart takes ~15s while the watermark model loads.

Where we are — honestly

Early access means early. Here's the truth table.

Live today

  • Marking core: C2PA + neural watermark (images)
  • Platform survival — survives a real Shopify upload — live-validated, not modelled
  • Durable recovery after metadata stripping
  • One-click compliance evidence pack (PDF)
  • Article 50 applicability engine + disclosure badges
  • Shopify app — label and register written at the point of publishing
  • Free exposure scan — chatbot and disclosure check, no account needed
  • Domain verification and per-customer registers, scoped and isolated

Be a founding design partner — shape it, free.

We're onboarding a small group of EU businesses that publish AI content. You get the compliance layer free while we build it around your workflow; we get the feedback that makes it right. Paid plans start at €29/month at launch.

No spam, no sharing. We'll reach out personally.

Questions, answered straight

Do I always need a visible "AI-generated" label?

No — and that's exactly why this is hard. The duty is conditional: deepfakes and AI text on matters of public interest need clear disclosure, but there's a carve-out where a human takes real editorial responsibility. RegisterWell applies the label where required and records the reasoning where it isn't — so you're covered either way.

My AI tools already mark their output. Doesn't that solve it?

It helps — and we read and respect those marks rather than replace them. But provider marks don't make your labeling decisions, don't survive every platform, and don't build your evidence trail across the dozen tools you actually use. That downstream layer is RegisterWell's job.

Is this legal advice? Do you guarantee compliance?

No and no — deliberately. RegisterWell performs the marking and labeling work and keeps the proof you did it. Our rule content is drafted against the Act and is under review by counsel, but legal responsibility stays with you, as it does with any compliance tool.

What does it cost?

Founding design partners: free while we build around your workflow. At launch, paid plans are expected from €29/month, rising with catalogue size, plus a one-off evidence pack for businesses that only need formal proof occasionally.

I'm not on Shopify. Is this for me?

The core is platform-agnostic — Shopify is simply where we're building the first one-click integration. If you publish AI content into the EU at any volume, join the list and tell us what your stack looks like.