[{"data":1,"prerenderedAt":378},["ShallowReactive",2],{"blog-en-medical-ai-safe-europe":3},{"id":4,"title":5,"body":6,"description":362,"extension":363,"meta":364,"navigation":371,"ogImage":372,"path":373,"seo":374,"sitemap":375,"stem":376,"__hash__":377},"updates/en/blog/medical-ai-safe-europe.md","Which medical AI is safe in Europe? What the American tools won't tell you",{"type":7,"value":8,"toc":344},"minimark",[9,26,50,53,56,61,64,69,77,80,84,87,96,99,102,116,119,122,125,128,132,135,138,141,144,148,151,154,157,160,163,166,169,172,176,179,182,185,188,192,195,198,201,204,207,211,214,217,220,223,226,230,233,236,239,242,246,258,267,276,285,294,298,307,311],[10,11,13],"key-takeaways",{"title":12},"In short",[14,15,16,20,23],"ul",{},[17,18,19],"li",{},"OpenEvidence withdrew from Europe last week; ChatGPT for Clinicians went live at the same time, available only to American doctors. Both show the same pattern.",[17,21,22],{},"Safety in medical AI breaks down into three layers: medical, data, regulatory. A tool that falls short on any one of these cannot be used in a European consulting room.",[17,24,25],{},"Ask three things of any AI tool before you use it: whether the use of information is transparent, where your data is stored, and how the tool relates to the EU AI Act.",[27,28,29,30,37,38,43,44,49],"p",{},"Last week something striking happened in the European AI landscape for doctors. OpenEvidence, the largest American AI search engine for doctors, ",[31,32,36],"a",{"href":33,"rel":34},"https://www.distilledpost.com/post/openevidence-withdraws-clinical-ai-platform-from-uk-and-europe-over-regulatory-uncertainty",[35],"nofollow","blocked European users itself",", citing the ",[31,39,42],{"href":40,"rel":41},"https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai",[35],"EU AI Act",". A few days earlier, OpenAI launched its new tool ",[31,45,48],{"href":46,"rel":47},"https://openai.com/index/making-chatgpt-better-for-clinicians/",[35],"ChatGPT for Clinicians",", but only for healthcare providers in the United States.",[27,51,52],{},"Two developments in one week with the same message: medical AI built for the US usually doesn't slot into Europe. It often needs a fundamental redesign. This isn't a matter of just copying it and rolling it out on the other side of the ocean.",[27,54,55],{},"\"So what?\", you might think as a doctor or other healthcare provider. But that is precisely why it's worth pausing here. Do these American solutions actually fit how we deliver care in Europe and our laws and regulations? In other words: are they safe to use here?",[57,58,60],"h2",{"id":59},"what-does-safe-ai-actually-mean-in-healthcare","What does 'safe AI' actually mean in healthcare?",[27,62,63],{},"'Safe' sounds simple, but for medical AI it means more than whether the tool gives a correct answer. For a doctor, safety is about at least three things. And on each of those points, a tool can still fall short.",[65,66,68],"h3",{"id":67},"_1-medical-content-and-appropriate-sources","1. Medical content and appropriate sources",[27,70,71,72,76],{},"Is the answer correct ",[73,74,75],"em",{},"and"," does it match the context in which you work? Many AI tools are developed on American healthcare information and context. You see this in the answers: what makes sense there does not always fit here.",[27,78,79],{},"In the Netherlands, the GP plays a central role, while care is organised differently in other countries. Prescribing behaviour, antibiotic use, referrals and patient communication also differ from country to country and region to region.",[65,81,83],{"id":82},"_2-data-processing","2. Data processing",[27,85,86],{},"It's not just about where data is stored, but above all what happens to that information. Is patient data used or deleted? Who has access? And as a doctor, can you explain to a patient or regulator what happens to that data?",[27,88,89,90,95],{},"This isn't only a theoretical risk. The Dutch ",[31,91,94],{"href":92,"rel":93},"https://demedischspecialist.nl/sites/default/files/2024-12/handreiking_gebruik_ai-chatbots_door_zorgverleners.pdf",[35],"Federation of Medical Specialists advises"," against using generic AI chatbots in direct patient care, precisely because of the lack of clarity around how data is processed and where it ends up.",[27,97,98],{},"The General Data Protection Regulation (GDPR) sets the framework, but the core is practical: you must retain control over patient information and be able to account for how it is processed. That covers consent, data minimisation, storage within the EU, and preventing data from quietly falling under foreign jurisdiction.",[27,100,101],{},"Under our regulations, as a doctor you must be able to account for:",[14,103,104,107,110,113],{},[17,105,106],{},"where patient data goes",[17,108,109],{},"who has access to it",[17,111,112],{},"how long data is retained",[17,114,115],{},"whether a patient can have their data deleted",[27,117,118],{},"Two practical requirements come on top of this.",[27,120,121],{},"First, as a doctor you are often the data controller for patient data. Any tool that processes that data must be able to enter into a valid data processing agreement. Many American tools don't offer this, or only in a form that doesn't meet what the GDPR requires.",[27,123,124],{},"Second, there is purpose limitation. Data you input may only be used for the purpose for which you process it: caring for your patient. If a tool also uses that same data to improve itself, that's a different purpose. Separate consent is required for that, and not every tool explicitly asks.",[27,126,127],{},"At Ask Aletta, personal patient data is automatically filtered out before a question is processed.",[65,129,131],{"id":130},"_3-laws-and-regulations","3. Laws and regulations",[27,133,134],{},"Does the tool comply with the rules that apply here? Think of the EU AI Act, but also existing medical legislation and standards for healthcare software, such as the Medical Device Regulation (MDR). That sounds logical and clear, but in practice this is the trickiest part.",[27,136,137],{},"Many of these rules are still new or being interpreted differently. What seems acceptable today may be judged differently a few months from now. That means that as a doctor, you don't just have to ask whether a tool 'is allowed' right now, but also whether it's built to stay within these rules as they're applied more strictly.",[27,139,140],{},"This is exactly where you see the difference between tools that were designed for the European context from the start and solutions that try to retrofit later. This is the point on which OpenEvidence got stuck.",[27,142,143],{},"For a doctor who wants to use a tool in practice, all three components must be in order. If one or more points aren't right, you can't responsibly use that tool, however good the answers may seem.",[57,145,147],{"id":146},"a-medical-chatgpt-doesnt-solve-it","A medical ChatGPT doesn't solve it",[27,149,150],{},"Does a medical version of ChatGPT solve this problem then? Not really. With the three points above, you can already answer that question.",[27,152,153],{},"Apart from the fact that ChatGPT for Clinicians isn't available in Europe at the moment, the tool has been developed for American healthcare. You see that in the sources the system is based on, the examples it uses, and the workflows it supports.",[27,155,156],{},"As a result, you get answers that make sense within the American context but don't naturally fit the European practice. So you can't use them one-to-one in a European consulting room.",[27,158,159],{},"The same goes for how data is handled. These tools are designed for a different context. The infrastructure and storage are often built around American healthcare systems.",[27,161,162],{},"For a Dutch or European doctor, that quickly means patient information is processed outside the EU, or falls under foreign jurisdiction. That directly touches medical confidentiality and the question of whether you can still account for what happens to that data.",[27,164,165],{},"That isn't a detail, but a precondition. If you can't make it clear where the data is and who has access, you can't use such a tool for patient information.",[27,167,168],{},"So the point isn't whether these tools are good, but what they're meant for. They've been developed for a healthcare system that differs from ours on multiple points. For European doctors, that simply means you can't use them one-to-one in the consulting room.",[27,170,171],{},"You see the same pattern with OpenEvidence. The fact that the tool has now withdrawn from Europe shows exactly what's at stake: if a system isn't designed for the European context from the start, it becomes difficult to meet those requirements after the fact.",[57,173,175],{"id":174},"three-flavours-of-medical-ai-tools-from-a-european-perspective","Three flavours of medical AI tools (from a European perspective)",[27,177,178],{},"If you look at the current offering as a doctor, you can roughly distinguish three kinds of medical AI tools.",[27,180,181],{},"The first category is generic AI chatbots/language models, such as ChatGPT from OpenAI, Claude from Anthropic and Gemini from Google. These are powerful models for general tasks, but they're not specifically developed for medical use. They don't operate within medical confidentiality and provide no guarantees about how patient data is processed. Partly because of that, they're unsuitable for use with patient information.",[27,183,184],{},"In addition there's specialised medical AI from the United States, such as ChatGPT for Clinicians and OpenEvidence. Often strong on content, but developed within a different healthcare system. You see that in the guidelines, the workflows and the way data is handled. This makes them not directly applicable in a European practice.",[27,186,187],{},"Finally, there's a growing group of European medical AI tools, including Ask Aletta itself. These have been designed for the European context from the start, with local guidelines and sources, European infrastructure and laws and regulations as a starting point. That makes them better suited for use in European healthcare practice.",[57,189,191],{"id":190},"what-europe-first-actually-means","What Europe-first actually means",[27,193,194],{},"Europe-first is being used more and more as a marketing term. But for medical AI, it should be about design choices: how is software set up and built from day one?",[27,196,197],{},"Does a tool work with European guidelines and sources, are those leading, and is it clear which sources are used for an answer? In the Netherlands, that means primary care guidelines, specialist protocols, medication information, and other reliable medical sources.",[27,199,200],{},"There's also where and under what jurisdiction the infrastructure runs. Where is your data? That determines whether, as a doctor, you can explain what happens to patient information.",[27,202,203],{},"The way a tool deals with laws and regulations also makes a difference. European rules like the EU AI Act and GDPR can't be 'added' afterwards. They have to be built into the design from the start.",[27,205,206],{},"And finally: transparency. As a doctor, you must be able to see what an answer is based on. Which guideline is it grounded in, and can you check that? Without that traceability, advice quickly becomes a black box.",[57,208,210],{"id":209},"what-this-means-for-a-doctor-choosing-an-ai-tool","What this means for a doctor choosing an AI tool",[27,212,213],{},"What does this mean concretely if you, as a doctor, are considering using an AI tool? It starts with a few basic questions.",[27,215,216],{},"What are the answers based on, and how current is that information? You want to know which guidelines and other reliable sources are used, and how often they're updated.",[27,218,219],{},"What happens to your data? Where is it stored, and under which jurisdiction? If that's not clear, or sits outside Europe, you can't use such a tool for patient information.",[27,221,222],{},"And finally: how does the tool relate to the rules that apply here? Think of the EU AI Act and existing medical legislation. If a vendor doesn't have a clear story on that, that's a signal the tool isn't ready for use in European practice.",[27,224,225],{},"These aren't theoretical considerations, but practical preconditions for use in the consulting room. They're exactly the points on which tools tend to get stuck in practice.",[57,227,229],{"id":228},"the-european-moment","The European moment",[27,231,232],{},"Something seems to be shifting in the market (and the world). Where American products long seemed self-evident, it's now becoming clear that they don't always fit within European healthcare practice.",[27,234,235],{},"With new regulation like the EU AI Act and parties having to limit their access, it's becoming visible that different requirements apply here. That has consequences for what you, as a doctor, can responsibly use.",[27,237,238],{},"With Ask Aletta, we deliberately choose to build for European practice: with local guidelines and sources, European infrastructure, and transparency and safety as starting points.",[27,240,241],{},"That means the choices discussed earlier in this blog have been built into the design of the tool from the start.",[57,243,245],{"id":244},"frequently-asked-questions-about-safe-medical-ai-in-europe","Frequently asked questions about safe medical AI in Europe",[247,248,251,255],"details",{"className":249},[250],"faq-item",[252,253,254],"summary",{},"Can I use ChatGPT for medical questions?",[27,256,257],{},"For general, non-patient-related questions, you can. But as soon as patient information is involved, it becomes problematic. You don't know exactly what happens to that data, and that means you can't properly account for it within medical confidentiality.",[247,259,261,264],{"className":260},[250],[252,262,263],{},"Is ChatGPT for Clinicians available in the Netherlands?",[27,265,266],{},"Not at the moment. The tool has been rolled out first to American healthcare providers. And even when it does become available in Europe, it remains a system that has been developed for American healthcare practice, with different guidelines and sources and a different way of handling data.",[247,268,270,273],{"className":269},[250],[252,271,272],{},"What does the EU AI Act say about medical AI?",[27,274,275],{},"Many applications in healthcare fall under the 'high risk' category. That means there are requirements around transparency, human oversight, and risk management. At the same time, part of the practical implementation is still being worked out, so it isn't always black-and-white what is and isn't allowed.",[247,277,279,282],{"className":278},[250],[252,280,281],{},"What European alternatives are there?",[27,283,284],{},"A growing range of medical AI tools developed specifically for the European context is emerging. Ask Aletta is one example, with a focus on Dutch guidelines, sources, and transparency in the source of answers. Other parties focus more on specific applications, such as speech-to-text, imaging, or triage.",[247,286,288,291],{"className":287},[250],[252,289,290],{},"What if I'm already using an American tool?",[27,292,293],{},"Then look critically at three things: what the answers are based on, what happens to your data, and how the tool relates to European regulation. If you don't get a clear answer on those, you're taking a risk that isn't always immediately visible.",[57,295,297],{"id":296},"try-it-yourself","Try it yourself",[27,299,300,301,306],{},"Curious how a European medical AI tool works in practice? You can ",[31,302,305],{"href":303,"rel":304},"https://app.askaletta.com/login/register",[35],"try Ask Aletta yourself"," and see if it fits the way you work.",[57,308,310],{"id":309},"sources","Sources",[14,312,313,319,325,331,337],{},[17,314,315],{},[31,316,318],{"href":33,"rel":317},[35],"OpenEvidence withdraws from EU/UK (DistilledPost, April 2026)",[17,320,321],{},[31,322,324],{"href":46,"rel":323},[35],"OpenAI launches ChatGPT for Clinicians (OpenAI, April 2026)",[17,326,327],{},[31,328,330],{"href":40,"rel":329},[35],"EU AI Act (European Commission)",[17,332,333],{},[31,334,336],{"href":92,"rel":335},[35],"Guidance on the use of AI chatbots by healthcare providers (Dutch Federation of Medical Specialists)",[17,338,339],{},[31,340,343],{"href":341,"rel":342},"https://www.autoriteitpersoonsgegevens.nl/themas/basis-avg/avg-algemeen/de-avg-in-het-kort",[35],"GDPR in brief (Dutch Data Protection Authority)",{"title":345,"searchDepth":346,"depth":346,"links":347},"",2,[348,354,355,356,357,358,359,360,361],{"id":59,"depth":346,"text":60,"children":349},[350,352,353],{"id":67,"depth":351,"text":68},3,{"id":82,"depth":351,"text":83},{"id":130,"depth":351,"text":131},{"id":146,"depth":346,"text":147},{"id":174,"depth":346,"text":175},{"id":190,"depth":346,"text":191},{"id":209,"depth":346,"text":210},{"id":228,"depth":346,"text":229},{"id":244,"depth":346,"text":245},{"id":296,"depth":346,"text":297},{"id":309,"depth":346,"text":310},"Which medical AI is safe for European doctors? A critical look at ChatGPT for Clinicians, OpenEvidence and what Europe-first really means.","md",{"date":365,"category":366,"author":367,"readingTime":368,"translations":369},"2026-05-07","blog","tijs.stehmann",8,{"nl":370},"medische-ai-veilig-europa",true,null,"/en/blog/medical-ai-safe-europe",{"title":5,"description":362},{"loc":373},"en/blog/medical-ai-safe-europe","D6FUg7B9WoxKVqPIaYVZngfOt5CaFiU_ibQlW9aQxcw",1778680911667]