The first AI-assisted IVF babies are now being born
It's a watershed moment for in vitro fertilization (IVF), which can be difficult and expensive. Doctors are hopeful AI can make it more effective.
In vitro fertilization (IVF) can be a lengthy, uncertain, and, for many, a frustrating process. Over the course of about two weeks, a woman hoping to conceive would inject herself with hormones so that her ovaries will produce multiple eggs. She would then have her blood drawn multiple times and have regular ultrasounds to track the progress of those eggs. Once she was deemed medically ready, she would be prepped for egg retrieval, a procedure requiring sedation, where her eggs would be collected then be sent to a lab and fertilized to create embryos.
After all that, there’s only a 50 percent chance an embryo can be transferred into the uterus, and it often takes several cycles of IVF before a successful pregnancy. For some women, IVF will never have the desired results despite the hefty price tag.
Still, the procedure is common. Around two percent of infants in the United States are born as the result of IVF and, in the four decades since the procedure was introduced, 10 million babies around the world have been born via IVF. But could IVF be easier and work better?
Several prominent fertility scientists and doctors tell National Geographic they see an exciting new era of IVF on the horizon thanks to advances in artificial intelligence. Recent developments in AI, including the development of tech that can detect sperm, grade embryos, and track embryos, aim to make the processes more accessible and effective for patients.
“The use of AI in reproductive medicine is truly at an incredible crossroads that is about to boom and revolutionize the way we practice reproductive health,” says Victoria S. Jiang, a physician and reproductive endocrinologist at Shady Grove Fertility in Atlanta.
Selecting healthier sperm
A team at Columbia University Fertility Center is expecting to soon see the first baby in the U.S. born via AI-assisted IVF. The patient became pregnant in March 2025 after 15 unsuccessful cycles of IVF.
Zev Williams, director of Columbia’s fertility center, has been using an experimental system called STAR (Sperm Tracking and Recovery) to aid in a critical point of failure in IVF: identifying the healthiest sperm. STAR, Williams says, uses AI to identify healthy sperm—meaning it’s alive, motile, and structurally intact—which would otherwise be undetectable to the human eye. After identifying the sperm, a robot removes the cells without the need for a centrifuge or other potentially damaging treatments. The process takes just milliseconds.
National Geographic History Magazine
Without an AI assist, an embryologist would need to look at a sample under the microscope, which could take hours and ultimately be fruitless. “In one case, experienced technologists searched through the sample for two days and found no sperm,” Williams says. “The sample was then run through STAR and STAR found 44 sperm in one hour.”
Unlike humans, AI can analyze millions of images without fatigue, detecting subtle patterns that the human eye can miss. “In severe male infertility, sperm can be so rare that even skilled embryologists may never see one,” Williams says. “It’s just beyond the capabilities of humans but very do-able using the special microfluidics chips (which are used to photograph and sort the sperm) we developed and the customized supercomputers and high-speed imaging that are part of the STAR system.” From here, humans take over—all subsequent IVF steps are performed using standard methods, Williams says.
The “early results,” Williams says, have been “very encouraging.” The system has repeatedly found sperm in samples from many men previously labeled “no sperm seen” and they’ve gone on to successfully create embryos.
(IVF revolutionized fertility. Will these new methods do the same?)
Keeping track of the healthiest embryos
Once the sperm meets the egg, the embryos sit in an incubator for five days. At that point, the embryo becomes a blastocyst and is graded based on the quality of its inner and outer cells. Both components receive a quality grade of A, B, or C. The two letters are then put together—so a blastocyst might be AA if it has the highest quality inner and outer cells or BC if it has mid-quality inner and low-quality outer cells.
To grade a blastocyst, an embryologist will look at time-lapse videos or single images of the embryos in addition to looking at them under a microscope. This process is subjective, time-consuming, and hasn’t changed much since the first IVF birth in 1978. But AI could change that: a 2024 clinical study published in Nature Medicine found that an AI system trained on data from over 115,000 embryos was nearly as good as humans at grading embryos.
After embryos are graded, genetic testing is often done to determine if an embryo is normal or abnormal. Usually, an embryologist hand tests each embryo to ensure it has the correct number of chromosomes. AI can help here, too.
In 2023, Nikica Zaninovic, Associate Professor of Embryology at Weill Cornell Medical College, and colleagues published a study in The Lancet that found that an AI algorithm can determine if an embryo has a normal number of chromosomes with about 70 percent accuracy, which is statistically significant. While traditional biopsy methods are around 90 percent accurate in ideal conditions, they can be less so in more difficult cases—for example, with mosaic embryos that contain a mix of normal and abnormal cells.
Researchers trained the AI system by feeding it pictures of both normal and abnormal embryos. From those images, the AI learned to predict whether embryos not included in the original dataset were normal or not. Since AI is evaluating images rather than the embryo itself, it’s not as invasive as the traditional biopsy, reducing the risk of damage to the fragile embryo.
The next step, Zaninovic says, is to add layers to the algorithm, so that it can help earlier in the process, like identifying unhealthy egg cells and predicting how many eggs a woman needs to ensure she freezes enough to have a successful pregnancy.
It also may help with the steps after grading and genetic testing, like assessing the thickness of the uterine lining as well as uterine receptivity after the embryo has been transferred. “Even after [preimplantation genetic testing], you still have miscarriages about 10 to 15 percent of the time,” Zaninovic says. “And that might not be related to the embryo itself, but it could be related to the overall patient wellbeing.”
Another of the challenges AI can help overcome has to do with avoiding potential mix-ups, including an embryo being mislabeled and implanted in a different patient. Embryos are so small—just 0.1 to 0.15 mm in diameter—that mix-ups have happened, though they are rare.
“If an embryo were accidentally moved to the wrong location within a dish, traditional systems would not catch it,” says Charles Bormann, director of embryology at the Massachusetts General Fertility Center. “AI systems are designed to track embryos throughout development and confirm identity at every step, ensuring that each embryo remains correctly labeled and matched to the right patient.”
The delicate process of making an embryo is central to at Conceivable, an IVF company based in Mexico. Alejandro Chavez-Badiola, an OBGYN and reproductive endocrinologist, uses a robotic arm guided by AI to automate over 200 steps in the IVF process, including keeping track of and transporting the sperm, egg, and embryos. For example, the tech helps identify “optimal” sperm, then a robotic system takes over, loading the sperm into an injection needle, positioning the egg, and then combining the two into an embryo. It does this while keeping tabs on the integrity of the egg, ensuring it’s not damaged during the delicateprocess.
Conceviable says the results speak for themselves. In 2024, the first baby was born using their proprietary tech. Since then, there have been 17 additional live births, according to Chavez-Badiola.
(Could this be the end as menopause as we know it?)
Experts urge caution
Despite the developments at research institutions and startups, there’s still reason to be wary of the limits of AI and its use in reproductive technologies.
Most of what happens in American IVF labs has no government oversight. Only certain components of the traditional IVF processes are regulated by the Food and Drug Administration (FDA), including some medical devices and genetic testing labs. Clinics can join the Society for Assisted Reproductive Technology (SART), which publishes success rates and ensures that member clinics are adhering to ethical guidelines, but membership is optional. Recently, the Centers for Disease Control and Prevention eliminated its IVF team.
This lack of oversight is true of AI-assisted IVF, too. Most of the current experiments are unregulated by the federal government except for AI systems that grade embryos. The FDA considers those medical devices and requires the agency’s approval. The only way to access some of the emerging assisted reproductive technology is in a research setting, like participating in one of the studies at a lab like Bormann’s or Zaninovic’s. (Startups like Conceivable that are outside of the U.S. are not subject to any of these regulations.)
There are also, experts note, numerous ethical questions around data privacy, algorithmic bias, and fairness.
According to a 2025 paper published in the Journal of Gynecology Obstetrics and Human Reproduction, many AI algorithms are hard to interpret by both doctors and patients, essentially functioning as "black boxes” because of their lack of transparency. This could be problematic, says Ali Abbara, associate professor in endocrinology at Imperial College London, because the AI could be making “illogical” decisions that the embryologists aren’t aware of. For example, if many normal embryos all came from the same incubator in the lab simply by coincidence, the AI model might wrongly assume that if an embryo is in a certain incubator, it’s more likely to be normal. Or it could start to factor in things like light, incorrectly grading an embryo as abnormal simply because an image is slightly darker. Additionally, the paper notes that because AI systems are trained on large, often sensitive, datasets there is potential for data breaches or misuse.
Researchers share concerns over the “black box.” “People are a bit worried about that because obviously in healthcare you want to understand it and trust it to some degree,” Abbara says. Transparent AI models that show their work can help solve this, Zaninovic says.
Jiang warns that using new tech before it’s ready for “primetime” is frequent in the field of reproductive endocrinology and infertility. She cites, for example, preimplantation genetic testing for polygenic risk scoring (PGT-P), a subtype of genetic testing that assesses everything from chromosomal and gene disorders like Down’s syndrome and sickle cell anemia to diabetes, heart disease risk, and intelligence.
“PGT-P uses a small sample of a few cells of the placenta layer of the embryo to be able to do genetic analysis for thousands of genes,” says Jiang, adding that it gives probabilities, not predictions, and accuracy cannot be guaranteed. “I think of it like a weather forecast—a 30 percent chance of rain doesn’t mean it will rain, and a five percent chance of heart disease doesn’t mean the person cannot get it.” But PGT-P testing is often sold to IVF patients as a guarantee of infant health or specific genetic traits, despite the lack of studies demonstrating its accuracy.
The analogy is true for AI, as well, and experts warn patients to be wary of big promises of guaranteed results. Still, researchers are excited about the future, even as they caution that AI should be a partner in care, not the sole provider. “AI can provide guidance and insights, but medical responsibility and patient autonomy should never be handed over to an algorithm,” Bormann says. Williams agrees: “AI must support, not supplant, the physician’s judgment,” he says. “As with any powerful innovation, responsible use, transparency, and clinical oversight are essential.”
