August 2026
Artificial intelligence is increasingly becoming part of modern fertility care. AI-supported embryo assessment, sperm analysis, treatment prediction, laboratory monitoring, patient communication and administrative automation are moving from experimental concepts into everyday fertility practice. But as the technology advances, so does regulatory scrutiny and 2 August 2026, important provisions of the EU Artificial Intelligence Act became applicable, including new transparency requirements.
For fertility clinics and technology suppliers, this is another important step towards a much more regulated environment for medical AI. The regulatory picture is broader than the EU AI Act alone and in the United States, the FDA has cleared at least one AI-enabled embryo-selection system for use in IVF, illustrating an important distinction: regulatory clearance may permit a product to be marketed for a defined intended use, but it does not mean that the technology is universally accurate nor compliant, clinically superior or free from the need for professional human judgment.
The US and EU approaches are not identical. FDA clearance generally focuses on whether a specific medical device may be marketed for a defined intended use under the applicable device pathway. The EU framework combines medical-device or IVD requirements with horizontal AI obligations, including requirements relating to risk management, data governance, transparency, human oversight, monitoring and documentation. Neither framework should be understood as a general endorsement of AI in fertility treatment.
Not all fertility AI tools are regulated in the same way
A scheduling system and an AI embryo-ranking algorithm clearly do not carry the same clinical risk. Administrative applications such as scheduling, billing and stock management will generally have a relatively limited AI Act risk profile, although they may still raise issues under data-protection, cybersecurity, employment and consumer-protection law.
Patient-facing AI requires greater attention. Clinics using chatbots, AI agents or automated patient-information systems should review whether patients are clearly informed when they are interacting with AI and whether the information provided is appropriately supervised. A system that provides general administrative information may present a different risk from one that gives personalised medical guidance or influences treatment choices.
At the other end of the spectrum are systems that influence clinical decisions, including:
- embryo grading and selection
- automated sperm assessment or selection
- implantation and live-birth prediction
- ovarian stimulation recommendations
- AI-assisted genetic interpretation
- laboratory automation.
These applications may potentially qualify as high-risk AI, particularly where the software is regulated as a medical device or IVD requiring third-party conformity assessment.
The fact that a system has received regulatory clearance or approval in another jurisdiction does not automatically determine its status in the EU. Nor does clearance necessarily establish that the system improves live-birth rates, is suitable for every patient population or should replace embryologist judgment. Those questions depend on the product’s intended purpose, evidence and clinical context.
AI within fertility may face multiple regulatory regimes
One of the most important issues for fertility technology companies is that compliance with the AI Act does not replace existing medical-device regulation. AI-based medical technologies may have to comply with both the EU Medical Devices Regulation or IVDR and the AI Act. In the United States, the same technology may also be subject to FDA device requirements, including the applicable classification, premarket pathway, labelling, quality-system obligations and post-market responsibilities.
This means developers need to consider not simply whether their technology uses AI, but its intended purpose, medical-device classification and role in clinical decision-making. Calling a product “decision support” or “embryologist assistance” does not in itself determine its regulatory status. Regulators will look at the product’s actual functionality, claims, instructions for use and reasonably foreseeable use.
The US experience provides a useful reminder that the wording of the intended use matters. An FDA-cleared embryo-selection product may be authorised for a specific function, such as assisting with the identification or ranking of embryos based on image analysis. That is not the same as a finding that the system can independently select the embryo most likely to result in a live birth, eliminate biological uncertainty or replace the treating specialist team.
FDA clearance should also be distinguished from FDA approval. The applicable pathway, predicate or review process, labelling and conditions of use all matter. A clearance decision is tied to the specific device and indications reviewed by the FDA. It should not be presented as a blanket certification of clinical superiority, universal applicability or guaranteed patient benefit.
Developers and clinics should therefore distinguish carefully between:
- regulatory authorisation to market a product
- analytical or technical performance
- clinical validity
- clinical utility
- evidence of improved patient outcomes.
These are related but separate questions. A product may demonstrate strong image-analysis performance without proving that its use improves cumulative live-birth rates. Conversely, a clinically useful tool may still produce uncertain or probabilistic outputs. Fertility outcomes are affected by many variables, including patient age, ovarian reserve, sperm factors, embryo biology, laboratory conditions, transfer decisions and chance.
Human oversight becomes increasingly important
Consider a simple IVF example.
An AI system ranks three embryos: 1 – 2 – 3
The embryologist independently ranks them: 3 – 1 – 2
Who decides?
The regulatory direction is not that clinicians should automatically follow AI recommendations. Rather, appropriately designed medical AI should allow trained professionals to understand the system’s purpose and limitations and to challenge, disregard or override its recommendations where appropriate.
Avoiding automation bias i.e. the tendency to trust a computer-generated recommendation simply because it comes from an algorithm, will therefore become an important part of clinical AI governance. Human oversight must be meaningful, not merely nominal. A clinician cannot realistically override a system if the output is unexplained, if the workflow discourages disagreement or if staff have not been trained to recognise failure modes.
Clinics should consider when an AI output must be reviewed, what information should accompany it, how disagreements are recorded and when use of the system should be paused. This remains true even where a system has been cleared by the FDA or authorised under another regulatory framework. Clearance generally reflects a defined regulatory assessment against a specified intended use but it does not transfer responsibility for the clinical decision from the embryologist or physician to the software.
Clinics have responsibilities too
Regulation is not solely the manufacturer’s problem. Under the AI Act framework, the technology manufacturer will normally be the provider, while the fertility clinic using the system will normally be the deployer. In the US as well as globally, clinics should also understand their responsibilities as users of an FDA-regulated device and ensure that the system is used consistently with its labelling, instructions and documented intended purpose.
Clinics should therefore increasingly consider:
- who is responsible for AI governance
- whether staff understand the systems they use
- appropriate human oversight
- whether input data are suitable
- monitoring unexpected results
- incident escalation
- documentation and logging
- GDPR/HIPAA compliance
- vendor responsibilities and contracts
- whether the system is being used only within its cleared, authorised or documented intended purpose.
AI literacy requirements have already applied since February 2025 under article 4 of EU AI ACT. Clinics should therefore ensure that clinicians, embryologists and other employees using AI have an appropriate understanding of both its capabilities and its limitations.
Patient communication also matters and should not be led to believe that an AI-generated embryo score is a guarantee of implantation, pregnancy or live birth. Where AI materially contributes to a treatment decision, clinics should consider how that contribution is explained, documented and incorporated into informed patient consent.
For a fertility clinic, this does not mean that everyone needs to become an AI expert. What it means in practice in a fertility clinic is that training should be proportionate to the employee’s role and the risk of the AI application.
| Person / function | Appropriate AI literacy |
| Reception/administration | Understand approved AI tools, confidentiality, patient data and when AI-generated information should not be relied upon |
| Marketing/customer service | Understand AI-generated content, hallucinations, patient transparency, GDPR and restrictions on entering patient data into public AI tools |
| Nurses/patient coordinators | Understand limitations of AI-generated patient information and when human/clinical escalation is required |
| Embryologists | Much deeper understanding of AI embryo/sperm assessment, validation, limitations, bias, automation bias and when/how to override recommendations |
| Doctors | Understand clinical validity, intended purpose, patient applicability, limitations and responsibility for AI-supported clinical decisions |
| Management/quality | Understand AI Act responsibilities, governance, risk classification, supplier responsibilities, monitoring and incident management |
| IT/DPO | Data processing, cybersecurity, GDPR, system integration, access controls and vendor/data governance |
In the US, clinics should also consider the product’s FDA labelling, user instructions, quality-system documentation and any applicable post-market reporting or monitoring expectations. A clinic should not expand a product’s use beyond the manufacturer’s documented intended purpose without carefully assessing the regulatory, clinical and liability implications.
Contracts with vendors should address data ownership and access, cybersecurity, software updates, model changes, validation after updates, audit rights, incident reporting, service continuity and responsibility for regulatory submissions. A system that changes over time may require renewed validation even if its name and user interface remain the same.
Data quality could become the fertility sector’s biggest AI challenge
An AI embryo-ranking model may perform very well in the population on which it was developed. But will it perform equally well for:
- older patients?
- donor-oocyte cycles?
- different ethnic populations?
- different incubator systems?
- different laboratory protocols?
- clinics in different countries?
Training and validation data, representativeness and bias are therefore likely to become increasingly important when assessing fertility AI.
For suppliers, large multicentre datasets and strong clinical validation could become significant competitive advantages. However, a large dataset alone is not enough as developers should also be able to explain how outcomes were defined, whether data were independently validated, how missing or poor-quality images were handled and whether performance was assessed across relevant patient and laboratory subgroups.
Clinics should be cautious about treating a regulatory clearance, a published accuracy figure or a vendor’s marketing claim as proof that a system will improve outcomes in their own setting.
They should ask whether the validation environment resembles their own laboratory, including imaging equipment, culture conditions, embryology protocols, patient mix and clinical workflow. Local validation may be particularly important where the system is sensitive to image quality, hardware or laboratory practice.
Developers should also plan for model drift. Changes in patient populations, laboratory procedures, imaging systems or clinical practice may affect performance over time. Monitoring should therefore continue after deployment rather than ending when the product reaches the market.
What should fertility organisations do now?
We recommend that clinics and fertility technology companies begin with a simple AI inventory. Identify every AI-enabled system currently being used and ask:
What does it do?
Does it interact directly with patients?
Does it influence a clinical decision?
Is it a medical device or IVD and has it been cleared, approved or otherwise authorised in any relevant jurisdiction, and for what precise intended use?
What regulatory pathway was used, and what limitations appear in the product labelling or instructions for use?
Who provides human oversight?
What evidence supports its performance?
Which data were used to validate it?
Has performance been assessed in the clinic’s own patient population and laboratory environment?
How are model updates controlled and revalidated?
Who is responsible if its recommendation is wrong?
How are patients and staff informed about the system’s role?
What happens if the system is unavailable, produces an implausible result or conflicts with professional judgment?
This exercise will often reveal that AI governance is no longer simply an IT issue. It is becoming a clinical governance, regulatory and management issue across both the EU and the US.
The opportunity remains significant
Stronger regulation should not be viewed simply as a barrier to innovation. AI has considerable potential to improve consistency, reduce workload, identify patterns humans may overlook and support better-informed fertility treatment.
The FDA’s clearance of AI-enabled embryo-selection technology demonstrates that fertility AI is moving into a more mature regulatory phase. At the same time, it also highlights the need to interpret regulatory decisions accurately. A cleared product may be legally marketable for a defined use without proving that it is superior to expert embryologists, appropriate for every patient or capable of predicting an individual patient’s outcome with certainty.
The EU AI Act adds another layer of accountability, but it does not turn probabilistic software into a clinical authority. The central question should remain whether the technology is appropriately validated, transparently presented and responsibly integrated into care.
The next phase of fertility AI will increasingly be about evidence, transparency and responsible implementation, rather than simply having the most sophisticated algorithm. For clinics, laboratories and fertility technology suppliers, the organisations that establish good AI governance now are likely to be best positioned for the regulatory environment ahead in both Europe and the United States.
Fertility Consultancy developed a framework illustrated below to support assessment of the AI situation at clinics, laboratories, gamete banks and technology suppliers. If interested knowing more reach out to us for a talk about its potential.
