T-minus … to EU AI Act Article 50 enforcement · 2 Aug 2026
EU AI Act · Article 50(1) · Chat widgets

Tawk.to and the EU AI Act: does your chat widget tell people it's AI?

Applies from 2 Aug 2026Fines up to €15M / 3% turnoverLast reviewed July 2026

Tawk.to is the internet's default free live-chat widget — and since AI Assist shipped, the same green bubble can be a human agent, Apollo AI Bot answering 24/7, or an AI that hands off mid-conversation. Visitors can't tell which from the outside: the widget looks identical either way. That ambiguity is the Article 50(1) problem in miniature — if Apollo answers first and nothing says so, your free chat widget quietly became an undisclosed AI system on an EU-facing site.

The rule itself is short. Article 50(1) of the EU AI Act requires AI systems that interact directly with people to be designed so users know they're dealing with AI — no later than the first interaction, in a clear and distinguishable way. The exception for cases where it's "obvious" is narrow: a natural-language customer-service bot doesn't qualify just because it has a robot icon. A line in your terms of service doesn't satisfy it either. And it's easy to enforce, because a regulator can simply open your site and start a chat.

Who's on the hook? Responsibility is split: Tawk.to carries provider-side duties for the system itself, but how the widget is configured and presented on your site is your deployment. If the disclosure setting exists and isn't enabled — or the notice is hidden — that gap is yours.

Where the disclosure lives in Tawk.to

In a typical Tawk.to setup, the surfaces that can carry (or fail to carry) the AI disclosure are:

Your Tawk.to disclosure checklist

Keep proof. On a complaint, an authority will want to see that the disclosure was there and how it was designed. Keep timestamped screenshots of the first-interaction state, your widget configuration, and a record of when each was last changed — that evidence file is the difference between "we comply" and "we can show we complied."

Watch out for

The classic tawk.to failure mode is inheritance: the widget predates AI on most sites, so nobody revisits its greeting when AI Assist gets switched on. The bubble that honestly meant 'chat with our team' for years starts meaning 'chat with an LLM' with zero copy changes — and because AI Assist is a per-workspace add-on any admin can enable, the gap can appear overnight without a deploy.

Note: vendor interfaces and setting names change. This page describes where disclosure surfaces typically live in Tawk.to as of July 2026 — verify the exact toggles in your own Tawk.to workspace and against Tawk.to's current documentation.

Common questions

We use tawk.to with human agents only — does Article 50(1) apply to us?

If no AI system interacts with your visitors, the chatbot-disclosure duty doesn't bite — human live chat isn't in scope. The catch is drift: the moment someone enables AI Assist (it's a dashboard add-on), the same widget becomes an AI interaction. Worth a periodic check rather than a one-time answer.

Where exactly do I add the disclosure in tawk.to?

Two layers: the AI Assist Instructions page (tells Apollo how to greet and behave — put the explicit AI line in the greeting behaviour there), and your widget's pre-chat/greeting copy. Belt and braces: if Apollo answers first, its first message should say what it is regardless of what the widget frame shows.

Check it in one scan.

DisclosureProof opens your site the way a regulator would, triggers your Tawk.to widget, and records whether the AI disclosure actually appears — with timestamped evidence either way.

Scan your site free