1. Regulatory focus
Article 50 of the EU AI Act does not simply say “disclose AI somewhere.” It separates obligations by use case: AI systems that directly engage in two-way interaction with people must inform users that they are dealing with AI, while systems generating or manipulating content must address machine-readable marking, detectability, and user-facing labelling. In its July 20, 2026 implementation guidelines, the Commission made clear that scope depends on concrete factors such as genuine bidirectional exchange, direct AI-to-person interaction, and whether the AI nature of the interaction is not already obvious to an average user. These obligations became applicable on August 2, 2026.
2. Business impact
For CPaaS providers, SaaS support teams, and cross-border digital platforms, the exposure is not limited to public website chat widgets. It extends to WhatsApp triage bots, in-app support assistants, ticket-routing copilots, login issue explainers, billing dispute flows, and other “semi-automated” conversation layers. If a company wraps AI responses under a human agent identity, or relies only on broad disclosure in privacy terms, that may fall short of Article 50. The risk increases further when AI-generated text is used to inform the public on matters such as public services, health, finance, consumer safety, or other public-interest topics. Providers outside the EU also need to pay attention: if the output is used in the EU, the obligations may still apply.
3. Operating recommendations
The practical move now is to treat AI transparency as a messaging-governance control, not a last-minute banner. Start by mapping every external conversation surface and classifying it: fully human, AI first response, AI-assisted human, or fully automated outbound flow. Then implement disclosure at the right points for in-scope use cases, including upfront notice, persistent in-conversation signalling where appropriate, handoff-to-human logic, and evidence logging. If your workflows also publish AI-generated text, summaries, or rich media, assess whether machine-readable marking and detection measures are needed. Where third-party models, bot platforms, or outsourced support vendors are involved, contracts should allocate responsibility for disclosure, labelling controls, audit logs, and change notifications.
Frequently Asked Questions
If our chat widget already says “virtual assistant,” do we still need extra disclosure?
Can we skip AI disclosure if the bot can eventually hand off to a human agent?
We operate support systems outside the EU. Do these rules still matter if we serve EU users?
Sources
- Commission publishes guidelines on transparency obligations for providers and deployers of certain AI systems
- Guidelines on transparency obligations for providers and deployers of AI systems
- Quick Facts: Transparency rules for AI systems
- Code of Practice on Transparency of AI-Generated Content
- Signing the Code of Practice on Transparency of AI-generated Content