PowerQuant provides technical compliance documentation — not legal advice.
On 20 July 2026, the European Commission adopted final guidelines on the implementation of the transparency obligations under Article 50 of the EU AI Act. On 2 August 2026, Article 50 itself entered into force. These obligations are no longer upcoming — they apply now.
What Are the Guidelines — and What Are They Not?
The Commission guidelines are non-binding. They provide practical guidance to competent authorities and to providers and deployers of AI systems. They do not replace the regulation text but clarify how the Commission expects Article 50 obligations to be implemented in practice.
Source: digital-strategy.ec.europa.eu — Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the AI Act (adopted 20 July 2026).
The Four Obligations Under Article 50
Article 50(1) — Chatbot and AI System Notification
Deployers of AI systems intended to interact directly with natural persons must inform those persons that they are interacting with an AI system — unless it is obvious from context. The guidelines clarify that "obvious from context" is a narrow exception, not a broad exemption. If your organisation uses AI-driven candidate communication, chatbots in recruitment, or automated conversational interfaces, the notification obligation applies.
Article 50(2) — Machine-Readable Marking of AI-Generated Content
Providers of AI systems generating synthetic audio, images, video, or text must ensure output is marked in a machine-readable format detectable as artificially generated. The guidelines recognise standards such as C2PA Content Credentials, Google SynthID, and cryptographic metadata as technical compliance paths. A visible watermark alone is insufficient.
For deployers: your vendor responsibility under Art. 50(2) must be documented through vendor attestation. The vendor marks technically — the deployer documents that marking occurs.
Article 50(3) — Emotion Recognition and Biometric Categorisation
Deployers of AI systems using emotion recognition or biometric categorisation must inform the affected natural persons. The obligation is absolute — there is no context exception as in 50(1). Systems that analyse candidate behaviour, sentiment, or video recordings from job interviews are typically in scope.
Article 50(4) — Deepfake Disclosure
Deployers and providers of systems generating or manipulating image, audio, or video content constituting a realistic deepfake must clearly disclose the artificial origin at the point of presentation. This is particularly relevant for AI-generated training video avatars, simulated interview scenarios, and similar HR-tech use cases.
What Did the Digital Omnibus Change — and What Did It Not?
A common misconception is that Regulation (EU) 2026/1744 (Digital Omnibus), which entered into force on 27 July 2026, postponed Article 50. It did not.
The Digital Omnibus contains a binding deferral to 2 December 2027 for the full obligation catalogue of standalone Annex III high-risk systems (technical documentation, conformity assessment, EU database registration). This deferral does not affect Article 50 transparency obligations, which remain applicable from 2 August 2026 unchanged.
Source: Regulation (EU) 2026/1744, published in the Official Journal of the EU on 24 July 2026.
Three Documents Your Organisation Should Have in Place Now
- AI inventory with Article 50 classification: A register of all AI systems in use, specifying which obligations (50(1)–50(4)) apply to each system and your organisation's role (provider or deployer).
- Vendor attestation for machine-readable marking: Written confirmation from your AI vendors that output is marked in machine-readable format, specifying the standard used (C2PA, SynthID, etc.).
- Disclosure texts for chatbots and AI interfaces: Concrete wording that informs users of AI interaction at session start or first contact.