EU AI Act Article 50 sets binding transparency obligations for AI systems and general-purpose AI (GPAI) models. The first enforcement window opens August 2, 2026. If your product uses AI-generated output, synthetic media, or an AI chatbot, the deadline to document your posture is now less than two months away.
What Juro does: Juro is a non-custodial website compliance scanner. It scans public-facing assets for disclosure gaps — including AI transparency markers required under Art. 50 — and produces signed, verifiable scan artefacts. It does not guarantee compliance; it shows you what's documented and what isn't.
What Does Article 50 Actually Require?
Art. 50 of the EU AI Act creates four distinct transparency obligations:
1. AI system disclosure to affected persons. Providers of AI systems that interact directly with natural persons — chatbots, automated advisors, customer service agents — must ensure those persons are notified they are interacting with an AI. The disclosure must happen before interaction begins, not buried in a terms page.
2. Synthetic content labelling. AI systems that generate synthetic audio, image, video, or text content must mark that content as machine-generated. The marking must be machine-readable and, where technically feasible, embedded in the output itself (not just a footer disclaimer).
3. Deep fake disclosure. Where AI-generated or AI-manipulated synthetic media depicts real persons, places, or events in a manner that could mislead the public, the deployer must disclose that the content was artificially created or manipulated. There is a narrow exception for legitimate artistic and creative expression.
4. Emotion recognition and biometric categorisation disclosure. Deployers of systems that infer emotion states or categorise individuals by biometric characteristics must inform the affected natural persons.
These are deployer obligations — meaning if your SaaS product deploys an AI feature, the obligation attaches to you, not to the foundation model provider.
What Is the GPAI Carve-Out?
General-purpose AI models — large language models, multimodal foundation models — have a separate transparency track under Art. 53 and Art. 55, which went live earlier. Art. 50 addresses the deployment layer: the product or service that puts a GPAI capability in front of an end user.
The carve-out that matters for most compliance teams: Art. 50(4) standard research exemption. Scientific research, development, and testing activities with no public deployment are outside scope. If your AI feature has shipped, this exemption does not apply.
A second carve-out applies to content that is obviously artistic or clearly satirical — the Act does not require labelling where the synthetic nature of content is self-evident from context. This is narrow. It does not cover marketing copy, product UI copy, or customer communications generated by AI.
Why August 2, 2026 Is the Relevant Deadline
The EU AI Act has a phased applicability structure:
- August 2, 2024 — Act entered into force
- February 2, 2025 — Prohibited practices (Art. 5) started applying
- August 2, 2025 — GPAI obligations (Title X) became applicable
- August 2, 2026 — Art. 50 transparency obligations become fully applicable for AI systems and deployed GPAI models
After August 2, 2026, national market surveillance authorities can act on Art. 50 enforcement actions against failures to disclose. Fines for deployers who fail to comply with transparency requirements can reach €15 million or 3% of global annual turnover, whichever is higher.
What Should Compliance Teams Do Before August 2, 2026?
The window before August 2, 2026 is enough to document your posture — it is not enough time to build new disclosure infrastructure from scratch. The practical priority order:
Map your AI touchpoints. List every AI-generated or AI-assisted user-facing output your product produces: chatbot responses, recommendation outputs, generated text, synthetic images. If you don't have this list, you can't assess Art. 50 scope.
Check your interaction disclosures. For each AI system that interacts with natural persons, confirm there is an unambiguous disclosure before or at the start of interaction — not a generic "we use AI" buried in an FAQ. The disclosure must be clear to the person in the moment.
Audit your synthetic content labels. If you generate synthetic media — images, audio, video — confirm you have a machine-readable marking mechanism. A visible watermark alone may not satisfy the machine-readable requirement; check what your generation tool provides.
Document your GPAI provider chain. If your product calls a foundation model API, document which model, which version, and which Art. 50 obligations you are responsible for as the deployer. Your foundation model provider's terms will not cover your deployment-layer obligations.
Date and sign your assessment. Regulators will ask when you assessed your Art. 50 posture. An undated internal assessment has less evidentiary weight than one with a signed date before the August deadline.
How Does This Interact With GDPR?
Art. 50 disclosure obligations are distinct from GDPR transparency requirements under Art. 13/14, but they stack. If your AI system processes personal data and interacts with data subjects, you need GDPR-compliant privacy notices and Art. 50 AI disclosure — both.
For companies also subject to DPDP (India) or DORA (EU financial sector), the documentation burden compounds further. Juro's scanner checks public-facing pages for GDPR, DPDP, and AI Act disclosure markers in a single scan, so you can see which layers are documented and which are not.
Frequently Asked Questions
Does Article 50 apply to internal AI tools used by employees, not customers?
Art. 50 applies to AI systems deployed with natural persons. For purely internal tools with no user-facing interaction output, the obligation is narrower — but if the tool's outputs reach customers (e.g., AI-drafted customer emails, AI-generated reports shared externally), the disclosure question re-enters. Assess by output destination, not just deployment context.
Does our foundation model provider handle Art. 50 compliance for us?
No. GPAI model providers have obligations under Art. 53/55 — technical documentation, training data summaries, copyright compliance policies. Deployers have Art. 50 obligations at the application layer. If your SaaS product uses an LLM API to power a user-facing feature, you are the deployer and the Art. 50 obligation attaches to you.
What counts as a machine-readable label for synthetic content?
The EU AI Act requires marking that is machine-readable and, where technically feasible, detectable. The European AI Office is expected to publish technical standards. Current best practice includes C2PA (Coalition for Content Provenance and Authenticity) metadata embedded in media files, visible watermarks plus EXIF/XMP metadata, and explicit API-level flags in output payloads. A disclaimer sentence in a footnote does not satisfy the machine-readable requirement.
We have an AI chatbot — what exactly does our disclosure need to say?
The Act does not mandate specific wording. The disclosure must make clear that the user is interacting with an AI system, before or at the start of the interaction, in a format they can understand. "You are chatting with an AI assistant" placed at the start of the chat interface is the baseline standard. Generic "powered by AI" branding in a product header is unlikely to satisfy the obligation.
What Juro Scans For on Art. 50
Juro checks public-facing pages for AI transparency disclosure markers: whether your chatbot interface carries an interaction disclosure, whether pages serving AI-generated content include a machine-readable marking indicator, and whether your privacy notice references your AI disclosure obligations. Juro documents what is and is not visible on your public surface — it does not assess your internal systems, model contracts, or provider agreements.
If you want to see which Art. 50 markers are present on your public pages before August 2, 2026, you can run a free scan at jurocompliant.com — no account required.
This post is for informational purposes and does not constitute legal advice. Consult a qualified legal adviser for obligations specific to your products and jurisdiction.
Free AI Act transparency scan — no account required
Juro scans your public-facing pages for AI transparency disclosure markers required under Art. 50: interaction disclosures, synthetic content indicators, and privacy notice coverage. It shows you what's documented and what isn't — before August 2, 2026.
Scan your site →