The short answer
Yes, if people in the EU use it. Since 2 August 2026, Article 50 of the EU AI Act requires that people are told they are interacting with an AI system unless it is obvious, that AI-generated content is marked in a machine-readable way, and that deepfakes are disclosed. Generative systems already on the market have until 2 December 2026 for the marking duty.
Key takeaways
- Article 50(1): chatbots and AI assistants must tell people they are interacting with AI, at the latest at the first interaction, unless it is obvious.
- Article 50(2): providers of generative AI must mark audio, image, video and text output as AI-generated in a machine-readable way. Systems already on the market have until 2 December 2026.
- Article 50(4): deepfakes, and AI-generated text published to inform the public, must be disclosed.
- Fines for breaching Article 50 reach €15 million or 3% of worldwide annual turnover, whichever is higher.
- California, South Korea, China and Texas have their own AI disclosure rules. SafeLaunch checks all of them in one scan.
On this page
01
Does the EU AI Act apply to my app?
The AI Act applies to providers (who build an AI system or put it on the market under their name) and deployers (who use an AI system in their product or business). It reaches companies outside the EU when their AI system is used in the EU or its output is used there.
Most apps built with AI builders are deployers: they call a model from OpenAI, Anthropic, Google or another provider and show the result to users. Article 50 puts duties on both roles, so a startup wrapping a model in a chatbot still has to make sure users know they are talking to AI.
| Your product | Article | Duty |
|---|---|---|
| Support chatbot, AI assistant, AI agent that talks to users | 50(1) | Tell people they are interacting with AI, unless obvious |
| Generates images, audio, video or text for users | 50(2) | Mark output as AI-generated in a machine-readable way |
| Uses emotion recognition or biometric categorisation | 50(3) | Inform the people exposed to it |
| Publishes deepfakes, or AI text on matters of public interest | 50(4) | Disclose that the content is AI-generated or manipulated |
02
When does Article 50 apply?
| Date | What applies |
|---|---|
| 2 Feb 2025 | Prohibited practices (Art. 5) and AI literacy for staff (Art. 4) |
| 2 Aug 2025 | Duties for general-purpose AI model providers |
| 2 Aug 2026 | Article 50 transparency duties |
| 2 Dec 2026 | End of the grace period for machine-readable marking (Art. 50(2)) for systems already on the market |
| 2 Dec 2027 | High-risk duties for Annex III uses (hiring, credit, education, essential services), as postponed by the Digital Omnibus |
| 2 Aug 2028 | High-risk duties for AI in products covered by Annex I |
03
How do I disclose a chatbot under Article 50(1)?
Article 50(5) says the information must be given in a clear and distinguishable manner, at the latest at the first interaction or exposure, and must meet accessibility requirements. In practice:
- Label the chat window with words, not only an icon: "AI assistant" in the header.
- Say it in the first message: "Hi, I'm an AI assistant. I can make mistakes. Ask for a person any time."
- Don't give the bot a human name and photo without the AI label. A persona is allowed; pretending to be human is the problem.
- Make the label readable by screen readers: real text with sufficient contrast, not text baked into an image.
- Keep it on hand-offs: if a human takes over, say so, and if AI drafts replies a human sends, consider saying that too.
Make this change without altering anything unrelated.
Our chat widget must tell users they are talking to an AI system before or at
the first message.
1. Add the visible text "AI assistant" to the chat header, next to the bot name.
2. Make the bot's first message start with: "Hi, I'm an AI assistant. I can make
mistakes. Type 'human' to reach a person."
3. Use real text (not an image), at least 4.5:1 contrast, and give the chat
region aria-label="AI assistant chat".
Done when:
- A new visitor sees "AI assistant" before sending a message.
- The first bot message states it is an AI.
- A screen reader announces the chat as an AI assistant.04
How do I mark AI-generated content under Article 50(2)?
Article 50(2) applies to providers of AI systems that generate synthetic audio, image, video or text. The output must be marked in a machine-readable format and be detectable as AI-generated, as far as technically feasible. If you only display a third-party model's output, check what your model provider already embeds, and don't strip it.
- Images and video: keep or add provenance metadata, such as C2PA Content Credentials, and don't remove metadata the model provider embeds.
- Audio: use your provider's watermarking where offered.
- Text: machine-readable marking of text is harder; record generation in your system and label AI text visibly where users could be misled.
- Visible labels help everywhere: a small "Generated with AI" tag also covers deepfake disclosure under 50(4) and similar rules in other markets.
05
Do other countries require AI disclosure?
Yes. If you sell beyond the EU, these rules from the SafeLaunch rulebook may apply too:
| Market | Law | What it requires |
|---|---|---|
| California | BOT Act (B&P §17941) | A bot used to sell something or influence a vote says it is a bot |
| California | SB 243 (from 1 Jan 2026) | Companion chatbots tell users they are talking to AI, have a self-harm protocol, and give known minors break reminders at least every 3 hours |
| California | AI Transparency Act, SB 942 (operative 2 Aug 2026) | Generative AI providers with over 1 million monthly users offer a free detection tool and embed disclosures |
| South Korea | AI Basic Act (from 22 Jan 2026) | Tell users in advance that a service uses generative or high-impact AI; label AI-generated outputs |
| China | AI Content Labelling Measures (from 1 Sept 2025) | Explicit and implicit (metadata) labels on AI-generated text, images, audio and video |
| Texas | TRAIGA, HB 149 (from 1 Jan 2026) | Prohibited AI uses; health care providers tell patients when AI is used in their care |
| Colorado | SB 26-189 (from 1 Jan 2027) | Tell consumers when automated decision-making influences consequential decisions; explain adverse decisions |
06
How SafeLaunch checks AI features
- Finds AI features in your pages and code, such as chat widgets and calls to AI model APIs.
- Checks for a disclosure a user would see before or at the first interaction.
- Reviews AI claims in your marketing for misleading statements under consumer law, such as the FTC Act §5 in the US.
- Applies each market's rule you select and cites the article, then writes the fix and a prompt for your builder.
- Adds an AI use policy to the Launch Fix compliance pack, which also supports the AI literacy duty in Article 4.
Run a free scan to see whether your AI features are flagged, or read how findings become fix prompts.
One free fix in your inbox each week
A new copy-paste prompt for Lovable, Cursor or Claude Code every week, with the rule behind it and how to check it worked. No spam, unsubscribe in one click.
By signing up you agree to get this weekly email. We send it with Kit, and you can unsubscribe at any time. What we share with Kit is in our privacy notice.
FAQ
Frequently asked questions.
Sources and further reading
This article is general information, not legal advice. SafeLaunch is automated software, not a law firm; a report records what was checked on one version and is not a certificate of compliance. Read the disclaimer.
From the founder
SafeLaunch is built by Raghav R Handa, a lawyer-turned-builder in New Delhi. He writes about this topic in plain English:
See what SafeLaunch finds on your product.
Paste a link and get your score, the most serious problems and what already passes, in about a minute.
Keep reading