AI embryo selection in 2026: what has changed, and what remains uncertain
Published and reviewed · Build AI Baby editorial summary
AI can make embryo assessment faster. Whether a particular tool improves the chance of taking home a baby requires a different kind of evidence.
This update reviews a 2026 professional committee opinion alongside a 2024 randomized trial. The trial is earlier evidence, not a new October 2026 breakthrough. This is educational writing, not a systematic review or personalized medical advice.
What the 2026 guidance adds
The American Society for Reproductive Medicine’s committee opinion describes potential roles for AI in IVF laboratories, including grading embryos and supporting workflow. It recommends caution: promising retrospective results need stronger prospective validation before widespread adoption.
The guidance identifies questions about transparency, data ownership and regulation. It calls for evaluations of outcomes including live birth, safety, time and cost. These are reasons to examine each tool individually, rather than treating all “AI baby” claims as the same technology.
Source: ASRM committee opinion, Fertility and Sterility, 2026;126:494–498
What a randomized trial actually measured
A 2024 trial randomized 1,066 patients across 14 clinics to iDAScore embryo selection or standard assessment by embryologists. Clinical pregnancy occurred in 46.5% of the AI group and 48.2% of the comparison group. The estimated difference was −1.7 percentage points, with a 95% confidence interval from −7.7 to +4.3 points.
The trial did not establish noninferiority: it could not rule out a disadvantage larger than its predefined five-percentage-point margin. This does not by itself establish that AI caused harm or that every AI model is inferior. Evaluation was much faster with AI, illustrating that laboratory efficiency and patient outcomes are separate measures.
The trial evaluated a particular system and selected patient population; its findings cannot automatically be applied to every clinic, model or family.
Original trial: Illingworth and colleagues, Nature Medicine, 2024 · PubMed record
Questions to take to a clinic
Our practical reading of these sources is to ask which model and version the clinic uses, whether it has been tested prospectively in comparable patients, which outcomes were measured, what it costs, and how an embryologist reviews the recommendation.
Embryo ranking does not edit DNA. This update does not establish the availability of a complete artificial-womb pregnancy or an ability to design a child’s traits.
How this update was prepared
We checked the professional guidance and the original trial’s publicly accessible report. We separated publication dates from our review date and retained the study’s uncertainty. The automatic PubMed feed provides discovery links; those records are not all reviewed articles.
Found an error or a newer result? Send the source through our contact form so it can be checked and this review corrected.
Back to research updates