The EU AI Act sets transparency rules for AI systems that interact with people or generate synthetic content. Avatar-based conversational AI falls squarely within its scope, so both we and you have obligations to meet.
This page is general product guidance, not legal advice. Your obligations depend on how and where
you deploy your product. Consult your own legal counsel to assess your specific case.
Risk Classification
EU AI Act Article 5 and Article 6 assign obligations by role. Beyond Presence is the provider of the underlying avatar and agent AI systems. You are typically the deployer, as you put them into use under your own authority and decide what they are used for.
Our AI systems are use-case agnostic, and the final application risk classification under the EU AI Act depends on your concrete use case and implementation scenario. Because we do not determine your intended purpose, classifying your application is your responsibility:
- Art. 5 lists prohibited AI practices, which may not be deployed at all. Among them are manipulative or deceptive techniques, exploiting the vulnerabilities of specific groups, social scoring, and emotion recognition in the workplace and in education.
- Art. 6, together with Annex III, defines which systems count as high-risk. For conversational avatars, the most relevant areas are employment and recruitment, education and vocational training, and access to essential private and public services. High-risk applications carry substantial additional obligations.
If you resell our services under your own name or trademark, substantially modify them, or change
their intended purpose, you may take on provider obligations yourself under
Art. 25. Assess this before launching a
white-label or heavily customized integration. Transparency Obligations
Article 50 of the EU AI Act sets out transparency rules highly relevant to avatars and agents:
- Article 50(1): Providers must ensure people are informed that they are interacting with an AI system, unless this is obvious from the context.
- Article 50(2): Providers of AI systems that generate synthetic audio, image, video, or text must mark the output in a machine-readable format so that it is detectable as artificially generated or manipulated.
- Article 50(3): Deployers of emotion recognition or biometric categorisation systems must inform the people exposed to them.
- Article 50(4): Deployers who generate or manipulate content that resembles real people, places, or events must disclose that the content is artificially generated.
- Article 50(5): Disclosures must be clear, distinguishable, and provided at the latest at the time of the first interaction or exposure, in an accessible way.
Article 50 applies from August 2nd, 2026. AI systems that were already on the market before that date have until December 2nd, 2026 to meet the machine-readable marking requirement under Article 50(2).
The sections below describe how Beyond Presence addresses each obligation, and where the responsibility sits with you.
AI Disclosure
Relevant under: Art. 50(1), 50(5)
All Beyond Presence Managed Agent calls have AI disclosure enabled by default. Participants are informed that they are speaking with an AI system before the conversation begins, without any configuration on your side.
If you use our white-label integration, customize the call UI, or build on speech-to-video, you are
responsible for adding an equivalent disclosure to your own product. Our default disclosure is part
of the standard Managed Agent experience, so replacing or restyling that experience removes it.
Watermarking
Relevant under: Art. 50(2), 50(4), 50(5)
Our text-to-video product in the Studio applies a visible watermark to generated videos by default, as required for artificially generated content under Art. 50(4).
We will soon also apply hidden, machine-readable watermarking to all our avatar model outputs as required under Art. 50(2). This will be invisible to your end users, and will be applied automatically, so no changes to your integration will be required.
Emotion Detection
Relevant under: Art. 5, 6(2), 50(3)
Beyond Presence Managed Agents with vision capability never perform emotion recognition. We do not infer emotions or intentions from facial expressions, voice characteristics, or other biometric signals, and we do not analyze protected attributes of end-user video. Because no emotion recognition or biometric categorisation takes place, the usage of vision capabilities does not entail further obligations. 
