EU AI Act Article 50 took effect on Sunday, August 2 — and if your product uses a chatbot, generates images or audio, or runs emotion recognition on users, you are legally required to disclose it. No grace period. Fines reach €15 million or 3% of global annual turnover. The obligation applies to any company whose product reaches EU users, regardless of where you are incorporated. If you haven’t audited your AI features against Article 50’s four disclosure categories, you are out of compliance today.
EU AI Act Article 50: The Four Categories That Require Disclosure #
Article 50 covers four categories of AI systems. First: AI systems that interact directly with individuals — chatbots, voice assistants, AI agents. Second: systems that generate or manipulate synthetic media, including audio, images, video, and text. Third: emotion recognition and biometric categorization tools. Fourth: AI-generated content on matters of public interest, including deepfakes. If your product touches any of these, disclosure is mandatory now.
The scope is broader than most developers assume. A customer service chatbot, an AI image generator, a mood-tracking wellness app, and an AI writing tool used for press releases are all covered. For chatbots and interactive agents, disclosure must happen at the first point of contact — not in terms of service, not in a linked FAQ, not in a modal that users can dismiss before reading. For synthetic content, providers must embed machine-readable markings and visible labels. For emotion recognition, individuals must be notified before analysis begins. For AI-generated content on public-interest topics — think AI-written news or political commentary — disclosure is required unless a human exercised substantive editorial review and assumes editorial responsibility for the content.
What You Actually Need to Build #
The most widespread implementation mistake is the metadata fallacy. Many teams are adding EXIF or XMP metadata tags to AI-generated images and calling it done. It isn’t. File-level metadata is invisible to users. It disappears when an image is screenshotted, re-saved, or shared through most platforms. Article 50 requires visible labels: embedded watermarks, persistent UI indicators, or visual markers that travel with the content beyond the originating platform.
For chatbots and agents, the “obvious AI” exception is narrower than companies want it to be. The EU AI Office’s July 20 guidelines clarify that the exception applies only when it is obvious “from the point of view of a natural person who is reasonably well-informed, observant and circumspect.” A chatbot named “AI Assistant” probably still requires explicit first-contact disclosure. A clearly cartoonish robot avatar in a toy company’s app might be obvious enough to qualify. When in doubt, disclose — the exemption is a narrow carve-out, not a general escape hatch. The disclosure must be persistent, not a one-time banner that users can skip.
For content generation tools that produce text on public-interest topics, human editorial review must be substantive — a single grammar pass does not constitute assuming editorial responsibility. Teams generating AI-written news summaries, financial analysis, or policy commentary need to define exactly what editorial review means internally, document it, and apply it consistently.
Related:[Colorado Rewrote Its AI Law: What Developers Need to Know]
This Applies to You Even If You Are Not in the EU #
Article 50’s extraterritorial reach follows the GDPR template: any provider, deployer, importer, or distributor placing AI on the EU market — or whose outputs are used within the EU — must comply. A startup headquartered in San Francisco with users anywhere in Europe is subject to Article 50 today. The fine cap is not reduced because you are not EU-based.
This is not theoretical. EU regulators have issued multi-million-euro GDPR fines against non-EU companies consistently since 2018. AI Act enforcement is expected to follow the same pattern. Additionally, California SB 942 — also effective in August 2026 — requires C2PA-compatible metadata in AI-generated media. Building for C2PA compliance covers a meaningful share of both the EU marking requirement and the California requirement simultaneously, reducing the implementation burden for teams serving both markets.
Deadlines and the Fast Track to Safe Harbor #
Most Article 50 obligations apply now — August 2. The one exception: existing generative AI systems already on the market before August 2 have until December 2, 2026 to implement machine-readable marking and detection mechanisms. All other obligations — chatbot first-contact disclosure, visible content labels, emotion recognition notices — have no extension.
The EU AI Office published a voluntary Code of Practice on AI-generated content transparency on July 22, 2026. Organizations that sign receive presumption of conformity and favorable enforcement treatment on marking and detection obligations. Signing the Code is the fastest route to regulatory safe harbor available now, before full technical implementation is complete. If your product generates substantial synthetic content and you are concerned about enforcement exposure while building the full marking pipeline, this is a practical interim step.
Key Takeaways #
- EU AI Act Article 50 is live as of August 2, 2026 — chatbot disclosures, synthetic content labels, and emotion recognition notices are mandatory now with no grace period
- Metadata-only implementations do not satisfy the visible-label requirement — labels must survive screenshotting and sharing to count
- The obligation applies globally to any product reaching EU users; being headquartered outside the EU is not a compliance defense
- The “obvious AI” exemption is narrow; when uncertain, add the first-contact disclosure rather than relying on the exception
- Sign the EU AI Office Code of Practice for presumption of conformity while building full compliance — existing systems have until December 2 for machine-readable marking only