2 August 2026 was the date we had been citing in our articles for months. It arrived. And two days later, on 4 August, Italy's Council of Ministers approved two legislative decrees implementing the AI Act.[2] The main headline: AI training becomes mandatory in Continuing Medical Education (CME). But there is also news that many commentators noted with relief: the most burdensome obligations for high-risk systems were postponed to December 2027 and August 2028 by the Digital Omnibus.[3] Let us clarify what is really in force today.
✓ In force today
Transparency obligations (Art. 50)
Any AI system interacting with patients must make its artificial nature evident. A medical chatbot must declare it is AI. A diagnostic support system must be presented to the patient as such.
Mandatory AI training in CME (Italian Decree 4 August 2026)
The Italian legislative decree of 4 August 2026 makes AI training mandatory in Continuing Medical Education. Details on credits, modalities and professional categories are being defined.
AI literacy for staff (Art. 4) — in force since 2 February 2025
Healthcare employers must ensure that staff using AI systems understand how they work, the legal limits and the risks. Training proportionate to role: developers, decision-makers, operational users.
Governance and sanctions (penalty regime applicable since 2 August 2025)
In Italy ACN and AgID supervise and sanction. The Italian penalty framework is graduated with lower maximum amounts than EU ceilings, but already enforceable for violations of obligations in force.
⏳ Postponed
Full obligations for autonomous high-risk systems → December 2027
Postponed by Digital Omnibus (May 2026). Covers Annex III systems, such as emergency patient triage; diagnostic software is a medical device and falls under the next item.
High risk integrated in medical devices → August 2028
AI systems that are medical devices (such as diagnostic software) or their safety components, already regulated by the MDR.
EU AI Act obligations for Italian physicians — status as of 11 August 2026
The Digital Omnibus point
The Digital Omnibus is a simplification package approved in provisional agreement between the Council and European Parliament in May 2026. It shifts full obligations for high-risk systems to 2027-2028. Note: it becomes legally effective only after publication in the EU Official Journal. Meanwhile, organisations are expected to prepare regardless — a compliance path for high-risk systems takes many months, and waiting until December 2027 is not a strategy.
A physician using an AI clinical documentation tool today should verify three things:
AI training in CME is not yet operational in its details — but it is already law. The physician who starts today using dedicated AI tools is not only improving their efficiency: they are building the practical competence that will soon become a formal CME requirement.
See also our articles on GDPR and medical AI in Italy and on how to choose a compliant AI tool.
Were the obligations for high-risk healthcare AI systems really postponed?
Yes, partly. The Digital Omnibus — a simplification package approved in provisional agreement between the Council and European Parliament in May 2026 — shifted full obligations for 'autonomous' high-risk systems (listed in Annex III, such as emergency patient triage) to 2 December 2027, and for AI systems that are medical devices, such as diagnostic software, or their safety components to August 2028. Note however: the postponement becomes legally effective only after the Digital Omnibus is published in the EU Official Journal. Obligations already in force (transparency, training, governance) were not postponed.
What does the transparency obligation mean for a physician using AI?
EU AI Act Article 50 — in force from 2 August 2026 — requires that any AI system interacting directly with people makes its artificial nature evident. In practice: a patient-facing AI assistant (chatbot on the practice website, intelligent booking system, automatic response) must openly declare it is an AI system and not a human. For the physician as deployer, this means verifying that all AI tools in use in their practice comply with this obligation.
What is AI training in CME and when does it become mandatory?
The Italian legislative decree approved by the Council of Ministers on 4 August 2026 makes AI training mandatory in Continuing Medical Education (ECM in Italian). At the time of publication, the government communiqué had not yet specified operational details: specific CME credits, course accreditation modalities, professional categories involved. These details will be defined by subsequent implementing decrees. AI training is likely to become a CME requirement.
Where does Clinovus AI stand with respect to the EU AI Act?
Clinovus AI is designed around AI Act principles. It is a documentation and sourced-information tool, not a medical device, and it never decides: the physician reviews, edits and signs every document structured from their dictation. Servers are in Switzerland, a country with an EU adequacy decision (2000/518/EC), and the model provider is contractually committed not to use the data for training. Clinovus AI is designed in line with the GDPR and the Swiss FADP.
Human oversight: the physician reviews, edits and signs every document structured from their dictation. Servers in Switzerland, a country with an EU adequacy decision; no clinical data sent to OpenAI, Google or other AI providers. Designed in line with the GDPR and the Swiss FADP.
Try free for 14 days →