Regulators can verify Article 50 the easy way: open your site and look. DisclosureProof does the same. It scans your pages, checks your chatbot disclosures and AI-content labels against the rules, and archives timestamped evidence your lawyer can put in a filing.
Scan your site free → See a sample report →
Free homepage scan, live now · evidence packs from €79, monitoring from €20/month · every paid plan starts with a 7-day free trial
Every Article 50 duty shows up on your live product, in plain view. That's why enforcement is simple, and why proof matters.
Any AI system that interacts with people must make that clear no later than the first interaction. A label buried in your terms doesn't count.
AI-generated audio, image, video, and text must be marked in a machine-readable, detectable format. Systems already on the market before 2 August 2026 have until 2 December 2026 for this one.
People exposed to these systems must be informed they're in operation.
Clear disclosure that content was AI-generated or manipulated, including news-style text published to inform the public.
Same four duties, different starting point. Browse by the chat widget you already run, or by the EU country your visitors are in.
Article 50, explained → Chat-widget guides → Country guides →
The scan mirrors what a supervisory authority (or opposing counsel) would do by hand, then keeps the receipts.
Paste a URL. We crawl your pages, trigger your chat widget, and sample your published media, all from the outside like any visitor. Nothing to install.
Every check maps to a specific obligation: disclosure present or missing, marking found or absent, labels rendered or not. The diagnosis is always free.
One purchase takes the scan to Business depth — up to 100 pages, all 9 checks, root causes — and seals the proof: timestamped screenshots, DOM captures, a hashed manifest, kept for a year.
Compliance drifts: widgets update, exports strip metadata, new channels launch unlabeled. Scheduled re-scans catch it before anyone else does.
Two of them produce a document about your intentions. One produces a record of what your site actually showed a visitor.
| Self-assessmentquestionnaires, checklists | Disclosure widgeta script that adds a banner | DisclosureProofan outside witness | |
|---|---|---|---|
| Where the answer comes from | What you tell it | What the vendor's own logs say | What your live site showed a visitor |
| Opens the chat widget like a visitor | – | – | Yes |
| Asks the assistant directly, and reads the answer | – | – | Yes |
| Gets past a cookie wall first | – | n/a | Yes |
| Covers all four Article 50 duties | Usually asks about them | Chatbot disclosure only | All four, graded separately |
| Works on a site you do not control | – | – | Any public URL |
| Evidence a third party can check | – | Vendor-held logs | Hashed, sealed, independently verifiable |
| Tells you when it silently changes | – | – | Scheduled re-scans with drift alerts |
| Says "compliant" about you | Often | Often | Never, and here is why |
Questionnaires ask you. Widgets promise for you. We check, and keep the receipts. A witness, not a widget. See the full comparisons → And how the EU's chat widgets actually behave is something we measured across the most-visited EU-facing sites and published, methodology and all: The State of AI Disclosure 2026 →
Laws differ; the check is the same: is the required disclosure there? DisclosureProof grades one scan against every rulebook that applies to you.
Chatbot disclosure, machine-readable marking, deepfake and AI-text labels. In force since 2 Aug 2026 for anyone serving EU users.
AI-generated news content disclosure for outlets operating in New York, alongside Article 50(4)'s public-interest text rule.
California SB 942 disclosures, Colorado's ADMT notice and record-retention duties (from 1 Jan 2027), Utah, and the rest of the patchwork.
Article 50 is the EU AI Act's transparency rulebook. It sets four disclosure duties: AI systems that interact with people must reveal they are AI; AI-generated image, audio, video, and text must be machine-readable as synthetic; anyone put in front of emotion-recognition or biometric-categorisation systems must be told; and deepfakes and AI-written public-interest text must be labeled. It binds both the provider that builds the AI and the deployer that puts it in front of users, and it applies from 2 August 2026.
Yes, it can. Article 50 follows your users, not your headquarters. Being a UK or US company is not an exemption. If your website, chatbot, or AI-generated content reaches people in the EU, the transparency duties can apply no matter where your company is registered, because the Act expressly covers providers and deployers outside the EU whenever the system's output is used in the Union. That is why US and UK businesses with any EU traffic are in scope.
It took effect on 2 August 2026. No, the 2025 “Digital Omnibus” did not move that date. The Omnibus, the EU's package to simplify the AI Act, pushed the high-risk system deadlines out to 2027–2028, and a lot of headlines shortened that to “the AI Act is delayed.” The Article 50 transparency duties stayed in place: they have applied since 2 August 2026, and the power to issue fines started the same day. The only part with a later date is machine-readable marking of AI content for generative systems already on the market before 2 August 2026, which has until 2 December 2026.
Up to €15 million or 3% of worldwide annual turnover, whichever is higher. Fines are issued by national market-surveillance authorities and can be levied since 2 August 2026; there is no separate enforcement grace period. Small businesses and start-ups are capped at the lower of those two figures rather than the higher, but the exposure is still real. And because Article 50 is not part of the high-risk regime, there is no conformity paperwork involved. The duty is only to disclose, and to be able to show that you did.
The user has to be told they are dealing with AI clearly and at the first interaction, not buried in your terms, not in a cookie banner, and not merely implied by a name like “Assistant.” In practice that means something visible in the conversation itself: an opening line such as “You're chatting with an AI assistant,” ideally alongside a persistent “AI” label on the chat window. There is a narrow exception when it is already obvious to a reasonable person, but a human-sounding support bot does not qualify. Whatever your chat shows on screen is what a regulator would judge.
Partly the vendor's, but not entirely. The Act divides duties between providers, who build the AI, and deployers, who put it in front of users. How the widget is configured on your site, under your brand, is a deployer responsibility that sits with you. Most major widgets already support an AI disclosure; it is often just switched off, worded too weakly, or hidden until someone clicks. “The vendor handles it” is worth confirming rather than assuming, which is exactly what a scan does.
Yes. Article 50 has no general small-business exemption. A two-person company running an AI chatbot for EU visitors carries the same disclosure duty as a large enterprise. The AI Act gives SMEs lighter supporting measures, such as priority access to regulatory sandboxes, and a lower maximum fine, but not a pass on transparency itself. Small teams are often more exposed, because the disclosure is missing for the simple reason that nobody turned it on.
Two separate duties. Article 50(2) covers synthetic media: AI-generated or AI-edited image, audio, video, and text, which must carry a machine-readable marking a detector can read. Article 50(4) covers what a person sees: deepfakes, and AI-generated text published to inform the public on matters of public interest, must be clearly labeled as artificial. An AI image, an AI voiceover, or an AI-written news explainer can each trigger these; a private draft you never publish generally does not.
Article 50 is unusually easy to check: a regulator or a competitor does not need your code, only your live site, and user or competitor complaints are a common trigger. Enforcement sits with national market-surveillance authorities, and the burden is on you to show you complied: authorities expect documented evidence (screenshots, configurations, timestamps), not verbal assurance. That is the gap DisclosureProof fills: it records what your site displayed, and when, so “the notice was there” is something you can prove rather than assert.
No. DisclosureProof is an informational scanning and evidence tool, not a law firm, and a scan is not a certification or a guarantee of compliance. It is built to surface disclosure gaps and preserve timestamped proof of what your site displayed; for how the law applies to your specific circumstances, use qualified counsel.
One URL in, findings out, graded against Article 50 with captured evidence sealed at scan time.
Scan your site free →