A doctor who uses a tool with artificial intelligence in the practice is, under European Regulation 2024/1689 (the AI Act), a «deployer»: someone who uses an AI system under their own authority. For them the basic obligations are few. Staff AI literacy applies from 2 February 2025, transparency towards individuals from 2 August 2026, and the heaviest obligations concern high-risk systems, which for products subject to medical device rules start from 2 August 2027. In Italy, Italian Law 132/2025 is added, which for healthcare says above all one thing: AI supports, the doctor decides.
Who is who: provider and deployer
The regulation divides the parties into roles. The provider is whoever develops an AI system, or has one developed, and places it on the market or puts it into service under their own name or trademark, for payment or free of charge (art. 3, point 3). The deployer is whoever uses an AI system under their own authority, except for personal and non-professional use (point 4).
If your practice management software has a dictation function that passes speech to a language model, the provider is whoever built and sold that function, and you who use it in the practice are the deployer. An AI system, under the regulation, is a machine-based system that infers from the input it receives how to generate output such as predictions, content, recommendations or decisions (art. 3, point 1). An automatic transcription that suggests fields has the characteristics of the definition; a comparison between two lists with fixed rules, without a model, is another matter.
What already applies: literacy and prohibited practices
Chapters I and II of the regulation apply from 2 February 2025. They contain two things that affect a practice.
AI literacy (art. 4). Providers and deployers take measures to ensure, to the extent possible, a sufficient level of AI literacy among their staff and anyone dealing with the operation and use of the systems on their behalf, taking into account technical knowledge, experience, training and the context of use. For a small practice this does not mean a certified course: it means that whoever dictates, whoever reads a draft and whoever answers patients knows what the tool does, where it goes wrong and what it must not be allowed to decide.
Prohibited practices (art. 5). The regulation prohibits, among others, the use of AI systems with subliminal or deliberately manipulative or deceptive techniques that materially distort a person's behaviour, and those that exploit vulnerabilities related to age, disability or social or economic situation. These are prohibitions concerning specific cases, but they are worth knowing if you use an automated assistant to talk to patients.
Transparency (art. 50): from 2 August 2026
Art. 50 sets out four obligations, two for providers and two for deployers. The European Commission has recalled that the new transparency rules take effect from 2 August 2026.
| Who | What | Article |
|---|---|---|
| Provider | An AI system that interacts with people must be designed so that they know they are dealing with an AI, unless it is obvious | 50(1) |
| Provider | Synthetic content (audio, images, video, text) must be marked in a machine-readable format | 50(2) |
| Deployer | Whoever uses a system that recognises emotions or categorises people on the basis of biometric data informs the people exposed | 50(3) |
| Deployer | Whoever generates or modifies images, audio or video that constitute a «deep fake» makes it known that they are artificial | 50(4) |
A «deep fake» is an AI-generated or manipulated image or audio or video content that resembles existing persons, objects, places or events and would falsely appear authentic (art. 3, point 60). For a practice the concrete question is about website and social media: an «after» image that AI has generated or retouched so that it looks real must be declared as such. The issue adds to the rules on healthcare advertising, discussed in what a doctor may write on their website and social media.
Another sentence of art. 50 matters for patient care: the information must be given in a clear and distinguishable manner, at the latest at the first interaction or exposure (par. 5).
High risk and medical devices
The most demanding obligations concern high-risk AI systems. Art. 6 identifies them in two ways.
The first, in paragraph 1, goes through products: an AI system is high-risk if it is intended to be a safety component of a product, or is itself a product, governed by the legislation in Annex I, and that product is subject to a conformity assessment by a third party. Annex I includes Regulation (EU) 2017/745 on medical devices, together with 2017/746 on in vitro diagnostics. By this route the obligations apply from 2 August 2027 (art. 113(c)).
The second, in paragraph 2, goes through the list in Annex III, which gathers specific uses in areas such as access to essential private and public services. Paragraph 3 provides for exceptions where the system does not pose a significant risk of harm, for example if it performs a limited procedural task or improves the result of a human activity already completed.
Whether a tool is a medical device depends on its intended purpose, not on the fact that it contains AI. The European guidance on software, discussed in is clinical records software a medical device?, explains how that is decided.
What the deployer of a high-risk system must do
Art. 26 lists the obligations. In summary:
- use the system in accordance with the instructions for use, with adequate technical and organisational measures;
- assign human oversight to people with the necessary competence, training and authority;
- if you control the input data, ensure that they are relevant and sufficiently representative of the system's purpose;
- monitor operation and inform the provider; if there is a risk or a serious incident, inform the provider and the authorities and suspend use;
- keep the automatically generated logs, to the extent under your control, for an adequate period and at least six months, unless otherwise provided by Union or national law, in particular on data protection;
- if you are an employer, inform workers before putting the system into service in the workplace;
- use the provider's information for the data protection impact assessment (art. 35 GDPR), where required.
Penalties
The regulation leaves the rules on penalties to the Member States (art. 99). For breaches of the deployer obligations in art. 26 and the transparency obligations in art. 50, a maximum of 15 million euros or 3% of annual worldwide turnover is provided, whichever is higher; for small and medium-sized enterprises the lower of the two amounts applies. Breaches of the prohibited practices in art. 5 reach 35 million euros or 7%. The Commission, in recalling the transparency rules, repeats the 3% cap and specifies that proportionality for SMEs and small mid-cap companies is taken into account.
Italian Law 132/2025
Italian Law no. 132 of 23 September 2025 sets out principles on artificial intelligence and declares in art. 1 that it is interpreted and applied in accordance with the European regulation. Two articles speak directly to those practising medicine.
Art. 7, healthcare. AI systems must contribute to improving the healthcare system while respecting the rights, freedoms and interests of the person, including on data protection. They may not select or condition access to services using discriminatory criteria. The data subject has the right to be informed about the use of AI technologies. The systems are a support in prevention, diagnosis, treatment and therapeutic choice processes, and the decision is always left to those practising medicine. The systems and data used must be reliable, periodically verified and updated to reduce the risk of errors.
Art. 13, intellectual professions. The use of AI is aimed solely at carrying out activities instrumental and supportive to the professional activity, with the intellectual work that is the subject of the service prevailing. Information on the systems used by the professional must be communicated to the recipient of the service in clear, simple and exhaustive language.
Art. 3 specifies that the law does not create new obligations compared with those of the European regulation for AI systems and for general-purpose AI models. The part that speaks to the doctor is that of informing the patient and the supporting role.
What to do in the practice
An afternoon's work, to be redone when something changes:
- List of the tools that use AI: dictation, website assistant, automated replies, image retouching, photo analysis
- For each one: who the provider is, where the data goes, what the contract says
- Who in the practice uses the tool and what they know about its typical errors
- A line in the patient information about the use of AI tools, in plain language
- No clinical decision delegated: the tool's output is a draft or a suggestion that you review yourself
- Images generated or modified by AI, if you publish them: declared as such
- Tools that could be medical devices: ask the provider for their written position
Where Fibonacci stands
In Fibonacci a language model appears in dictation: it transcribes speech and suggests fields, saves nothing on its own and formulates no diagnoses or therapy advice. The artificial intelligence page lists what never happens and where the data runs; how dictation works in detail is in dictation during the consultation. Fibonacci declares when you are talking to an AI; machine-readable marking of generated content is not there yet.




