AI systems in healthcare have long ceased to be research projects alone. Chatbots answer questions, assistants structure information, and generative systems produce text for patients, employees or the public. The transparency obligations in the European AI Act take effect in August 2026. This turns a matter of good communication into a concrete governance task.
People must know when they are interacting with a machine
The AI Act takes a risk-based approach. According to the European Commission's current overview, people must, for example, be informed when they are interacting with an AI system such as a chatbot, unless this is already obvious. The Commission dates the start of these transparency rules to August 2026.
AI-generated content should also be recognisable as such. More extensive labelling obligations apply to certain content, for example deepfakes or texts published on matters of public interest.
This clarity is particularly relevant in healthcare. Patients and their relatives must be able to assess whether a response was generated automatically, what limits the system has and when a human takes over.
Where transparency can quickly be absent in day-to-day practice
Most risks do not arise from a single spectacular AI system, but from many small applications:
- A chatbot answers questions about services, appointments or preparation.
- An assistant drafts emails, website text or patient information.
- An internal system summarises documents or suggests next steps.
- A speech or translation tool supports communication.
- Generative AI creates content that is then published without a discernible review.
It is often not documented where such systems are already being used. Transparency can then neither be designed consistently nor reliably demonstrated.
Four tasks for healthcare organisations
Inventory AI applications
Organisations should first identify where AI is used, including embedded functions in standard software. The relevant factors are the purpose, target group, types of data, responsible parties and whether the output reaches patients or the public directly.
Label touchpoints clearly
For patient-facing chatbots and assistants, it must be clear that a machine is involved. The notice should not be hidden in general terms and conditions, but appear where the interaction begins.
Organise human escalation
Transparency alone is not enough. Users need a clear route to a responsible person, particularly for medically sensitive, misunderstood or urgent enquiries. A system must not imply certainty that is not supported organisationally.
Adapt publication processes
Review, approval and evidence should be defined for AI-assisted text and media. This applies to website content, patient information, campaigns, social media and internal communication. The organisation remains responsible, not the model being used.
Not every AI system is automatically high-risk — but each needs a clear purpose
The AI Act distinguishes between different levels of risk. Whether a system is considered high-risk depends on the product, function and context of use. A general service chatbot requires a different assessment from an AI component that influences safety-relevant medical decisions.
For this reason, a blanket classification is of limited use. Organisations need a structured process that considers the use case, data, people affected, potential harm, human oversight and regulatory classification together.
Conclusion: Trust is created through visible responsibility
The transparency obligations are not an obstacle to meaningful use of AI. They create a minimum standard for how organisations demonstrate responsibility. Patients and employees must be able to recognise when AI is involved, what role it plays and where human decision-making authority begins.
Primary sources and scope of validity
- European Commission: AI Act – Regulatory framework, accessed 11 August 2026. It substantiates the risk-based framework and the Commission's stated start of the transparency rules in August 2026.
- Regulation (EU) 2024/1689 on artificial intelligence, official EU legal act. It is authoritative for assessing the specific obligations and applicability.