The EU AI Act and Clinical AI: What Applies Now
In 2026 the EU moved its most demanding AI rules back by more than a year. A lot of people read the headline, filed AI compliance under 2027, and moved on.
That is the mistake worth avoiding, because the delay covers less than it appears to, and the parts already in force are the parts most likely to touch a clinic this year. None of the obligations were softened. Only the clock moved.
Here is what applies today, what is coming, and what to ask before you buy anything.
What changed, and what did not
The AI Act is Regulation (EU) 2024/1689. In 2026 it was amended by a package known as the Digital Omnibus on AI, Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026 and in force from 27 July.
The Omnibus pushed back the dates on which high-risk obligations begin to apply. It did not remove any category of obligation, and it did not change the Medical Device Regulation or the In Vitro Diagnostic Regulation, which run on their own schedule regardless.
The dates that matter
| Date | What applies |
| 2 Feb 2025 | Prohibited practices under Article 5, and the AI literacy duty under Article 4. Both already in force. |
| 2 Aug 2025 | Obligations for providers of general-purpose AI models, governance rules and national penalty regimes. |
| 2 Aug 2026 | Transparency obligations under Article 50. National market surveillance authorities begin enforcing the AI literacy duty. |
| 2 Dec 2026 | End of the transitional period for marking AI-generated content, for generative systems placed on the market before 2 August 2026. |
| 2 Dec 2027 | High-risk obligations for stand-alone Annex III systems. Moved from 2 August 2026. |
| 2 Aug 2028 | High-risk obligations for AI in regulated products under Annex I, which includes AI-enabled medical devices. Moved from 2 August 2027. |
| 2 Aug 2030 | High-risk systems used by public authorities, including those already in service. |
Two of those dates are near. The content-marking transitional period closes on 2 December 2026, and the AI literacy duty has been enforceable since August 2026.
Are you a provider or a deployer?
This is the distinction that decides which obligations are yours, and clinics routinely get it wrong by assuming none of it applies to them.
A provider develops an AI system and places it on the market under its own name. That is the vendor.
A deployer uses an AI system under its own authority in the course of its activity. That is the practice, the hospital, the clinic.
Deployers of high-risk systems carry their own obligations under the Act, and they are not trivial. They have to use the system in line with the provider’s instructions, assign human oversight to people with the necessary competence, training and authority, and give those people the support they need. Where they control the input data, they have to make sure it is relevant and sufficiently representative for the system’s intended purpose.
They also have to monitor the system against its instructions for use, tell the provider when something looks wrong, and suspend use and notify the market surveillance authority where the system may present a risk. Serious incidents are reported to the provider first, then to importers, distributors and authorities. Automatically generated logs within the deployer’s control are kept for at least six months unless other law says otherwise.
Two further duties are easy to overlook. Before a system is used in the workplace, employers inform workers’ representatives and the affected workers. And for Annex III systems that make or assist decisions about people, those people are told the system is being used on them.
Which clinical AI is high risk
Broadly, there are two routes in.
Annex I, through product law. AI that is a medical device, or a safety component of one, where the device requires third-party conformity assessment. These sit inside the existing MDR and IVDR machinery. Deadline 2 August 2028.
Annex III, stand-alone uses. Health-related uses that are high risk in their own right, including systems that infer health status from physical or biological signals, systems that determine access to healthcare, and systems that triage or prioritise emergency care. Deadline 2 December 2027.
A single product can fall under both. The boundary is genuinely difficult in places, which is why the Commission has been working on classification guidance, and why this is a question for your regulatory adviser rather than a blog.
The duty that already applies to almost everyone
Article 4, AI literacy, has applied since February 2025 and has been enforceable by national market surveillance authorities since August 2026. It applies to providers and deployers regardless of risk class, which means a practice using an AI scribe is in scope even though that tool may never be classified as high risk.
The Omnibus rewrote this obligation in a helpful direction. The original duty was to ensure a sufficient level of AI literacy. The amended duty is to take measures supporting its development. You are not expected to guarantee a particular level of competence for any individual, and there is no certificate to obtain.
The duty covers your own staff and others using the systems on your behalf, including contractors. In practice, the reasonable response is modest: a short induction for anyone using an AI tool clinically, covering what it does, what it does not do, where it fails, and who to tell when it misbehaves. Keep an internal record that it happened.
Transparency: Article 50
These obligations have applied since 2 August 2026 and are not limited to high-risk systems. In outline, they require that:
- People are told when they are interacting with an AI system rather than a person.
- AI-generated or manipulated content is marked in a machine-readable format.
- Those exposed to emotion recognition or biometric categorisation systems are informed.
- Deepfake content is disclosed.
For a clinic, the first two are the live ones. Any patient-facing chat or triage assistant falls squarely in the first. The second matters if you publish AI-generated material, and it carries the 2 December 2026 date for generative systems that were already on the market before August 2026.
The Commission published guidelines on Article 50 in July 2026, and there is a voluntary code of practice on AI-generated content which the Commission and the AI Board treat as adequate evidence of compliance.
Why the delay is not a reason to stop
Three reasons.
The obligations did not change. Risk management, data governance, technical documentation, logging, human oversight, accuracy and robustness, post-market monitoring: all still there, all still required, just later. Building that into a product takes longer than the extension.
Some of it is already enforceable. Transparency and AI literacy are not waiting for 2027.
Your customers will ask before the regulator does. Procurement questionnaires move faster than enforcement. A hospital buying a clinical AI tool in 2027 will ask how it meets the Act, and the honest answer in 2027 depends on what was built in 2026.

What to ask a vendor now
- Are you a provider under the AI Act, and do you consider this system high risk? Under which annex, and on what basis?
- If it is not high risk, what is the reasoning, and is it written down?
- What Article 50 transparency applies to this product, and how is it implemented in the interface my patients see?
- What human oversight does the system assume, and what does it expect the clinician to check?
- What logs does the system generate, who controls them, and how long are they kept?
- What happens if the system behaves unexpectedly? Who do we tell, and what is your incident process?
- What information will you give us to support a data protection impact assessment?
- Are you certified to a quality system such as ISO 13485, and how does your AI Act work sit inside it?
Those eight questions will tell you more about a vendor’s maturity than any marketing page.
Where LiteStream sits
Our design position has been the same since before the Omnibus, and it is the one the Act’s human oversight provisions assume. The software flags and drafts. The clinician decides and approves. Visual analysis surfaces potential concerns for a clinician to evaluate rather than producing a conclusion, and every clinical note is reviewed and approved by the clinician before it is saved.
We are not going to tell you in a blog post how LiteStream will be classified under the Act. That determination belongs with our regulatory advisers and the relevant authorities, and any vendor that gives you a confident one-line answer on a marketing page is telling you something about how seriously they take it.
LiteStream is in working beta, and we are adding trial partners now.
Frequently asked questions
Has the EU AI Act been delayed?
Parts of it. Regulation (EU) 2026/1744 moved the application of high-risk obligations to 2 December 2027 for stand-alone Annex III systems and 2 August 2028 for AI in regulated products such as medical devices. The obligations themselves were not changed, and other parts of the Act already apply.
What already applies to a clinic using AI?
The prohibited practices and the AI literacy duty have applied since February 2025, with national enforcement of AI literacy since August 2026. The Article 50 transparency obligations have applied since 2 August 2026.
Is an AI scribe a high-risk system?
Not automatically. High-risk status depends on the classification rules, and a documentation tool may fall outside them. That does not put it outside the Act, because the transparency and AI literacy duties are not limited to high-risk systems.
Does the AI Act replace the Medical Device Regulation?
No. The Omnibus did not change the MDR or the IVDR, which continue to apply on their own schedule. An AI-enabled medical device can be subject to both frameworks.
What is the difference between a provider and a deployer?
A provider develops the system and puts it on the market under its own name. A deployer uses it under its own authority in the course of its activity. A clinic is usually a deployer, and deployers of high-risk systems carry their own obligations.
This article summarises published legislation and regulatory guidance and is not legal advice. Regulatory dates and interpretations change. Check the current position and take advice before relying on anything here.
Choosing clinical AI and want straight answers on oversight and compliance? Request early access and join the clinician waitlist.
Sources
- Regulation (EU) 2024/1689 (the AI Act), consolidated text. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
- European Commission AI Act Service Desk. Article 26: Obligations of deployers of high-risk AI systems. https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-26
- AI Act: obligations in force and delays under Omnibus 2026/1744. https://www.actuia.com/en/news/ai-act-obligations-in-force-and-delays-under-omnibus-2026-1744/
- Freshfields. The final Digital Omnibus on AI: key amendments to the AI Act. https://www.freshfields.com/en/our-thinking/blogs/technology-quotient/eu-ai-act-unpacked-34-the-final-digital-omnibus-on-ai-key-amendments-to-the-a-102nber
- Gibson Dunn. EU AI Act Omnibus Agreement: postponed high-risk deadlines and other key changes. https://www.gibsondunn.com/eu-ai-act-omnibus-agreement-postponed-high-risk-deadlines-and-other-key-changes/
- EU AI Act deadlines for healthcare AI and medical devices. https://www.healthseed.vc/insights/eu-ai-act-healthcare-deadlines



