August 2026
New EU AI Act provisions have become operational from 2 August 2026.
Artificial intelligence is entering reproductive medicine at increasing speed. Much attention has focused on AI embryo selection, but AI is increasingly relevant before an embryo even exists including sperm analysis, gamete assessment, donor programmes, predictive analytics and potentially donor-recipient matching.
For sperm and egg banks, donor programmes and fertility clinics, this raises an important question:
How should AI be used responsibly and in legal compliance, when decisions involve donated human reproductive cells?
AI and gamete selection
Computer vision and machine-learning systems can analyse sperm and oocyte images and potentially identify characteristics that are difficult or time-consuming for humans to evaluate consistently.
AI could therefore increasingly support:
• sperm assessment (e.g. motility, concentration, morphology)
• sperm-selection technologies
• oocyte assessment
• gamete quality prediction
• laboratory quality control
• donor-treatment outcome prediction.
The potential benefit is considerable: more standardised assessments, less subjective variation and potentially better use of large datasets.
But gamete selection also illustrates why reproductive AI requires careful governance.
A recommendation is not a decision
One principle is particularly important as AI should support appropriately qualified professionals rather than silently replace their judgement.
If an AI system recommends one sperm sample, oocyte or treatment strategy while an embryologist or clinician reaches another conclusion, there must be clarity about who makes the final decision and how the AI recommendation should be interpreted.
European medical-AI regulation increasingly emphasises meaningful human oversight and protection against automation bias. That principle is particularly relevant in donor conception, where decisions may ultimately affect not only the recipient patient but also donor-conceived people.
Donor data requires special attention
AI systems are only as reliable as the information on which they are trained. For donor programmes this can involve highly sensitive information, including:
• medical history
• genetic information
• phenotype
• family history
• infectious-disease screening
• gamete characteristics
• treatment outcomes.
AI does not remove existing GDPR responsibilities. Organisations must therefore understand which data an AI system uses, where those data are processed, whether they are being used to train future models and which organisation is responsible for the processing.
Bias can enter donor programmes quietly
AI promises greater objectivity, but algorithms can reproduce biases contained within their training data. A gamete-assessment model trained on a narrow population or using data from a limited number of laboratories may not perform equally well across other populations, laboratories or treatment protocols.
The same concern becomes even more important if AI is eventually used for donor-recipient matching or predictive selection. Donor programmes should therefore ask technology providers:
Which population was the algorithm trained on?
Has it been independently validated?
Does performance vary between patient or donor groups?
Can staff override the recommendation?
How is performance monitored after implementation?
What changes under the EU AI Act?
Important EU AI Act provisions have now become operational from 2 August 2026 with transparency requirements applying to certain AI interactions and generated content. Organisations using patient-facing AI therefore need to consider whether users are appropriately informed when interacting with an AI system. AI used for medical purposes may face considerably stronger requirements.
Automated sperm analysis or selection, for example, may fall within both medical-device legislation and the AI Act depending upon the technology’s intended purpose and classification. For high-risk medical AI, areas such as validation, data governance, robustness, human oversight, documentation and post-market monitoring become increasingly important.
Donor programmes should not assume that purchasing an approved AI product is sufficient. Staff need to understand what the system predicts, and what it does not. A LAB tech or embryologist using AI-assisted sperm assessment should understand the system’s intended purpose, limitations and validation population and know when human judgement should override the algorithm.
AI literacy is therefore becoming part of good clinical and laboratory governance.
The donor sector should engage early
AI has significant potential within gamete donation, where it can assist laboratories improve consistency, reduce manual workload and extract useful insights from datasets that are increasingly difficult for humans to analyse alone.
But donor conception deserves particularly careful implementation as the donor sector deals not simply with technology and patients, but with donors, recipients and donor-conceived people whose interests can extend decades into the future.
The question should therefore not simply be:
“Can AI help us make this decision?”
It should also be:
“Should AI influence this decision, what evidence supports it, and where must human responsibility remain?”
Those questions should be at the centre of responsible AI adoption in gamete donation and gamete-selection with AI analysis and ranking of both embryos and gametes, including the potential to reduce subjective evaluation.
Fertility Consultancy developed a framework illustrated here to assess the AI situation at clinics, companies and gamete banks. If interested knowing more reach out to us for a talk about its potential.
