Research: Reproductive Selection and Polygenic Embryo Screening

July 2, 2026

Scope: eugenics-adjacent algorithmic decision-making with reproductive implications, AD 2016 to AD 2026. Six threads, each documented with named actors, dates, and a source where the claim can be re-verified. Where a company makes an efficacy claim its own peer-reviewed literature does not support, the claim is recorded as the company’s, not adopted as fact. Where a journalist or advocate labels a company “eugenicist” — or the industry calls the same product “empowering choice” — the label is attributed to whoever made it, not adopted here.

1. Algorithmic child-welfare decisions

1.1 Allegheny Family Screening Tool — design and history

Allegheny County, Pennsylvania, became the first U.S. jurisdiction to put a statistical risk score on every child-abuse hotline call. The Allegheny Family Screening Tool (AFST) was built by Rhema Vaithianathan (Centre for Social Data Analytics, Auckland University of Technology) and Emily Putnam-Hornstein (then USC, now UNC Chapel Hill) under a contract with the county. The model returns a score from 1 to 20 estimating the two-year probability that a child named in a hotline call will be placed in foster care. Screeners see the score before they decide whether to send a caseworker (Allegheny County DHS — AFST landing page; Methodology V2 white paper, May 2019).

In the January 2019 V2 release the county doubled the share of referrals flagged “mandatory high-risk” from ~11.5% to ~22% (rising to ~24% in production), and in October 2019 lowered the age cutoff for the low-risk protocol from twelve to seven years old (Methodology V2 PDF; Wichita State PPMC summary, 2024).

Patrick Ball’s Human Rights Data Analysis Group (HRDAG) obtained AFST training data and ran an independent disparate-impact analysis. HRDAG reported that households where a person with a disability was a resident received elevated scores, and that access to county-run mental-health services added up to three points on the twenty-point scale — meaning, in HRDAG’s account, that the act of seeking help became a feature of the abuse-risk prediction (HRDAG analysis, 22 June 2023; ACLU, “How Policy Hidden in an Algorithm is Threatening Families”).

Sally Ho and Garance Burke of the Associated Press reported on 31 January 2023 that the Department of Justice Civil Rights Division was investigating AFST under Title II of the Americans with Disabilities Act after three complaints from civil-rights attorneys, including a lawyer for two parents with disabilities whose daughter was removed. The probe, as reported, targets whether AFST systematically scores households with disabled residents higher and whether the county’s reliance on the score amounts to disability discrimination (PBS NewsHour syndication of the AP report; WESA (Pittsburgh NPR), 31 January 2023). DOJ has not, as of this page’s compilation, issued a public finding letter; the matter is an open investigation, not a proven violation.

Erin Kissane and others published “The Allegheny Family Screening Tool’s Overestimation of Utility and Risk” in Logic Magazine (October 2023), documenting that the original 76% accuracy claim referred to AUC under a re-referral proxy outcome, not actual abuse confirmation (Logic Magazine).

1.2 The Hackney case

Andrew Hackney (borderline intellectual functioning, post-stroke neurocognitive disorder) and Lauren Hackney (mild intellectual disability, anxiety) took their then-eleven-month-old daughter to a doctor because she would not take her bottle. A medical professional filed a report with the county’s Office of Children, Youth and Families (CYF); the AFST score was elevated; the child was removed and placed with a foster family. Per court filings and hearing testimony, a CYF caseworker testified that the Hackneys “should not have unsupervised visits because their disabilities prohibited them from safely caring for the daughter” and acknowledged the agency had not arranged a parental-capacity evaluation. The Hackneys sued in federal court, naming the county, three caseworkers, and three supervisors; a federal judge cleared the case to proceed in April 2026. These are allegations pending in litigation cleared to proceed, not adjudicated fact (TribLive; WPXI, April 2026).

1.3 Replication and abandonment

Oregon deployed an Allegheny-derived screening algorithm in 2018, attempted a “fairness correction” against documented racial disparity, then announced on 19 May 2022 it would withdraw the tool by end-June 2022 and replace it with the Structured Decision Making model, citing equity concerns (NPR, 2 June 2022; The Imprint).

The UK’s London Borough of Hackney ran an Early Help Profiling System (EHPS) pilot with private vendor Xantura on a child-protection profiling system, then dropped it after concluding the system did not “realise the expected benefits” and that data-quality problems made continued investment unjustifiable. The council declined a Freedom of Information request for contract financial details (EachOther; Community Care).

ACLU reporting documents that child-welfare agencies in at least twenty-six U.S. states plus Washington D.C. have considered algorithmic screening tools; at least eleven have deployed them. Named deployments outside Allegheny include Los Angeles County (California) and Larimer, Arapahoe, and Douglas counties (Colorado) (ACLU, “Family Surveillance by Algorithm”; ACLU report PDF, 28 September 2021).

1.4 The pattern

The case-law and complaint record draws a consistent picture: parents who receive disability benefits, mental-health services, or Medicaid home-and-community-based services produce elevated risk scores in proportion to how visible they are to county data systems. The signal the algorithm captures is contact with public services, not abuse. Bauserman v. UIA (Michigan MiDAS), Perdue v. Gargano (Indiana FSSA), and Legal Aid of Arkansas v. ARChoices establish the same pattern in benefits-determination contexts; the Allegheny line extends it to family integrity — the same logic deciding whether a parent gets to keep the child she already has.

2. Polygenic embryo screening companies

The hard line in this section: separate what each company claims its product does from what peer review supports. Company foundings, funding rounds, named backers, published prices, and product launches are drawn from company disclosures, filings, and press. Every efficacy figure — IQ uplift per embryo, disease-risk-reduction percentage — is recorded as the company’s or an advocate’s claim. Peer review (the American College of Medical Genetics and Genomics called clinical use “premature” in 2023; Karavani et al., eLife 2022; critics Lior Pachter and Sasha Gusev) contests within-family predictive validity, so those figures are attributed, never asserted as delivered results. The “eugenics” / “consumer eugenics” label, where it appears below, is the reporter’s or critic’s, attributed — not this page’s voice.

2.1 Orchid Health (San Francisco)

Noor Siddiqui, a Thiel Fellow and former Stanford AI / genomics researcher, launched Orchid in 2021 after watching her mother go blind from retinitis pigmentosa. Orchid’s distinguishing technical claim is whole-genome sequencing of an IVF embryo from a five-cell biopsy — Orchid states this covers ~99% of the embryo’s three billion base pairs, versus the ~0.25% covered by conventional PGT-A screening. Price published as $2,500 per embryo sequenced, on top of the IVF cycle cost (CNBC, December 2023; Orchid product page).

Orchid announced that the first child conceived via its full-genome screen, named Jaffe, was born in late 2023, and that the post-birth genome matched the pre-implantation sequence — the company’s own announcement (Orchid announcement).

Orchid’s $4.6M seed round named Anne Wojcicki (23andMe co-founder), Brian Armstrong (Coinbase co-founder), and Fred Ehrsam (Coinbase co-founder) as backers. The Washington Post (Pranshu Verma and Nitasha Tiku, 16 July 2025) profiled Silicon Valley’s “consumer eugenics” wave with Orchid as the flagship; the “consumer eugenics” framing is the Post’s (Washington Post; Fortune, “Silicon Valley sets its sights on building the perfect baby,” 29 November 2025).

UC Berkeley computational biologist Lior Pachter published a technical critique of Orchid’s polygenic-score claims on his Bits of DNA blog (April 2021), arguing the within-family predictive value is bounded by the heritability ceiling and that the marketing language (“over 1200 conditions”) obscures the gap between detecting rare Mendelian variants (well validated) and predicting common complex disease risk (poorly validated for selection) (Bits of DNA).

2.2 Genomic Prediction / LifeView (North Brunswick, NJ)

Genomic Prediction Inc. was co-founded in 2017 by Stephen Hsu (Michigan State University theoretical physicist) and Nathan Treff (statistical geneticist), and markets the LifeView test for preimplantation genetic testing for polygenic disorders (PGT-P) (Wikipedia: Stephen Hsu, with citations to MIT Technology Review and GenomeWeb; MIT Technology Review, Antonio Regalado, “The world’s first Gattaca baby tests are finally here,” 8 November 2019).

Hsu’s 2014 Nautilus essay “Super-Intelligent Humans Are Coming” argued that selection across roughly ten embryos could shift expected child IQ upward by ~15 points with adequate predictive power. The ~15-point uplift is Hsu’s own projection, contingent on predictive power he concedes does not yet exist — recorded as his argument, not as an established result. In interviews with Razib Khan and Dwarkesh Patel he has repeatedly framed embryo selection as the near-term route to that effect, and told The Guardian “accurate IQ predictors will be possible if not in the next five years, the next 10 years certainly” (Razib Khan Substack; Dwarkesh Podcast).

Hsu has stated in interviews that Genomic Prediction does not return polygenic IQ scores to LifeView customers — only outlier flags for predicted IQ below 75 (intellectual disability). Critics including Sasha Gusev and Lior Pachter note that the underlying scoring infrastructure is the same (GEN interview).

In June 2020 the Center for Genetics and Society documented an MSU graduate-student-union campaign (#FireStephenHsu) citing Hsu’s blog and tweets, his promotion of figures including Holocaust deniers, and alleged conflict-of-interest disclosures around his Genomic Prediction board seat. MSU President Samuel L. Stanley announced Hsu’s resignation from the vice-president-for-research position on 19 June 2020; Hsu retained his tenured faculty appointment. The resignation date and retained tenure are on the record; the underlying characterizations are the campaign’s allegations, attributed to it (Center for Genetics and Society).

2.3 Heliospect Genomics / PolygenX

Heliospect Genomics, a US-incorporated company, obtained UK Biobank access in June 2023; its principal investigator of record on the application was Alexandros Giannelis, and the stated research purpose was development of “advanced techniques on new genetic data” for “complex traits.” The application did not disclose embryo screening or IQ prediction as a commercial application (HOPE not hate, “Access to UK Biobank”).

Hannah Devlin (Guardian science editor) and HOPE not hate published a joint undercover investigation (sting conducted November 2023; published 18 October 2024): an undercover reporter posing as a prospective customer was offered the PolygenX service for $50,000 to screen up to 100 embryos. Heliospect CEO Michael Christensen (Denmark) appeared on a video call and described the IQ-uplift target as approximately six IQ points per selected embryo over a naturally conceived sibling. The $50,000 price and the video-call offer are the investigation’s on-record findings; the ~6-point IQ uplift is Christensen’s sales claim, not a demonstrated result (The Guardian, “US startup charging couples to ‘screen embryos for IQ’,” 18 October 2024; HOPE not hate, “The Superbaby Factory”).

UK Biobank issued a public statement disputing the report’s framing, asserting that its investigation found “no evidence of misuse” because Heliospect used only publicly accessible summary statistics. The statement did not address the application-purpose-mismatch finding (HOPE not hate page above; Wikipedia: UK Biobank).

Polygenic embryo screening is prohibited in the UK under the Human Fertilisation and Embryology Act 1990 (as amended); it is unregulated at the federal level in the US. Heliospect markets to UK and EU customers from US soil (Alliance Vita summary).

2.4 Nucleus Genomics (New York)

TechCrunch reported Nucleus Genomics’ $14M Series A (January 2025) and founder Kian Sadeghi’s public launch of “Nucleus IQ,” described on X as “the first intelligence score based on your DNA.” The Series A and launch are on the record; the “intelligence score based on your DNA” characterization is Sadeghi’s own marketing, not a validated predictive capability. Sasha Gusev (Dana-Farber computational geneticist) publicly challenged the test’s predictive validity, and Sadeghi posted a lengthy defense (TechCrunch, “Controversial genetics testing startup Nucleus Genomics raises $14M Series A,” 30 January 2025; TechCrunch, “Genetics testing startup Nucleus Genomics criticized for its embryo product,” 6 June 2025).

Beverly Hills Fertility announced a clinical partnership with Nucleus Genomics for embryo polygenic scoring on its IVF panel, the first US clinic chain to publicly integrate Nucleus’s pipeline (Femtech Insider).

2.5 Herasight

Herasight, named for the Greek goddess Hera, emerged from stealth in 2025. It offers polygenic risk reduction for childhood and adult disease, plus predictions for height, BMI, longevity, and IQ, and claims 20–44% disease-risk reduction when selecting among five embryos. The company’s existence, product scope, and method paper are on the record; the 20–44% disease-risk-reduction figure and the cognitive-ability predictor are Herasight’s own marketing claims, not peer-validated results. Herasight released a method paper, “ImputePGTA,” on medRxiv (7 November 2025) describing accurate embryo genotyping from ultra-low-pass sequencing (Herasight Substack, “Announcing the world’s most powerful genetic predictor of cognitive ability”; medRxiv preprint; SynBioBeta profile).

Rob Stein’s NPR report “These labs say they can screen embryos for specific traits. Should you?” (6 May 2026) treats Herasight and Nucleus alongside Orchid and Genomic Prediction as a single emerging product category, and quotes Hank Greely (Stanford) and Françoise Baylis (Dalhousie) on the regulatory void (NPR).

An independent geneticist critique, “Embryo selection company Herasight goes all in on eugenics” (Theinfinitesimal Substack, 2025), contends Herasight’s IQ-PGS methodology repackages the same within-family limitation Pachter identified for Orchid; the “eugenics” framing is the author’s (Theinfinitesimal).

2.6 Industry context

Antonio Regalado’s “The race to make the perfect baby is creating an ethical mess” (MIT Technology Review, October 2025) surveys the polygenic-screening landscape including all five companies above (MIT Technology Review).

The American College of Medical Genetics and Genomics (ACMG) has not approved polygenic embryo screening for clinical use; its 2023 position paper called such use “premature.” A January 2022 Karavani et al. paper in eLife concluded that screening embryos for IQ and other complex traits remained premature (Science magazine summary).

3. Genetic-database property and class actions

3.1 Henrietta Lacks Estate settlements (2023–2026)

The settlements below record resolutions, not admissions of wrongdoing — the Lacks suits settled on undisclosed or confidential terms, and no admission was made public.

The Estate of Henrietta Lacks v. Thermo Fisher Scientific Inc. (D. Md., filed October 2021) alleged that Thermo Fisher had profited from HeLa cell lines taken without Henrietta Lacks’s consent in 1951. It settled on undisclosed terms on 1 August 2023, one day after what would have been Henrietta Lacks’s 103rd birthday (Health News Florida (WUSF); Science magazine; Seeger Weiss litigation page).

The Estate of Henrietta Lacks v. Novartis Pharmaceuticals Corp. reached a confidential resolution in February 2026; a companion case against Viatris Inc. resolved in March 2026 (AFRO American Newspapers; C&EN).

Predating the litigation, NIH and the Lacks family agreed in 2013 that NIH-funded researchers sequencing HeLa lines must submit the sequence data to a controlled-access database, and that a review board including Lacks family members evaluates new access applications — the template for family-as-stakeholder in genetic-data governance (Queen’s Gazette).

3.2 23andMe Chapter 11 and the state-AG opposition

23andMe Holding Co. entered Chapter 11 (D. Mo., E.D., 23 March 2025) with the express aim of selling the genetic database — ~15 million customer profiles — as an asset. Anne Wojcicki resigned as CEO; the company’s market capitalization had collapsed from ~$6B post-SPAC to under $50M (HIPAA Journal summary).

Attorneys general of more than two dozen US states filed objections to the proposed sale (June 2025), arguing that genetic data is a categorically different asset class than ordinary bankruptcy collateral. Pennsylvania AG Sunday joined the multi-state action; the “different asset class” argument is the AGs’ attributed legal position (Pennsylvania AG press release).

At the 4 June 2025 auction, TTAM Research Institute — a nonprofit set up by Anne Wojcicki in May 2025 — submitted a winning bid of $305M, beating Regeneron Pharmaceuticals’ $256M offer (23andMe investor-relations release; NPR, 30 June 2025).

U.S. Bankruptcy Judge Brian Walsh approved the sale on 30 June 2025 (close 14 July 2025). California, Kentucky, Tennessee, Texas, and Utah remained on record as actively opposed at the time of approval. TTAM committed to honoring the existing privacy policy, establishing a consumer-privacy advisory board, and producing annual reports for state AGs on request (U.S. News; Foley Hoag privacy analysis, July 2025).

Public Citizen’s July 2025 policy brief argued the case proves the U.S. Bankruptcy Code requires amendment to exclude biometric and genetic data from saleable bankruptcy assets (Public Citizen).

Separately, a $30M settlement resolved class actions arising from the October 2023 credential-stuffing breach that exposed ~6.9M users’ relative-finder data (InformationWeek).

3.3 Florida DNA Privacy Act and ongoing PGT-A class actions

Florida HB 1189 (signed June 2020, effective 1 July 2020) was the first state law in the US prohibiting life, disability, and long-term-care insurers from requesting or using genetic-test results in underwriting (University of Miami Law Review).

Florida HB 833, the Protecting DNA Privacy Act (effective 1 October 2021), is codified at Fla. Stat. § 760.40 and § 817.5655. The statute declares “the genetic information of the person from whom it is extracted is the exclusive property of that person to control” and attaches criminal penalties (third-degree felony for unlawful disclosure) to each instance of intentional collection, retention, analysis, or sale without express consent (2025 Florida Statutes § 760.40; Hunton analysis; Inside Privacy briefing).

Class actions alleged that Natera Inc.’s noninvasive prenatal screening (NIPT) tests were marketed as reliable for rare-disorder detection when their false-positive and false-negative rates undermined the marketing; the suit settled for $8.25M under consumer-protection statutes in Florida and elsewhere. These are plaintiff allegations resolved by settlement, not proven misconduct — a settlement is not an admission (Top Class Actions; Claim Depot).

Fertility Bridge tracks an expanding wave of plaintiff filings (2024–2026) against multiple PGT-A providers alleging misrepresentation of euploidy-screen accuracy and the role of false-mosaicism results in the disposal of viable embryos; Florida is a recurring venue because of HB 833’s “exclusive property” provision. These are allegations (Fertility Bridge).

4. Reproductive surveillance — biometric and identity stack

4.1 Aadhaar and reproductive-benefits gatekeeping (India)

India’s conditional cash-transfer scheme for pregnant women, Pradhan Mantri Matru Vandana Yojana (PMMVY), conditions ₹5,000–11,000 benefit payments on Aadhaar authentication (Government of India, Ministry of Women and Child Development).

Economist Jean Drèze (Ranchi University) and co-authors (Drèze, Khera, and Somanchi, 2021) documented that ~29% of eligible women in their sample faced disruptions in PMMVY benefit receipt attributable to Aadhaar authentication failures — mismatched names, missing bank–Aadhaar linkage, biometric-read failures (Tribune India summary; The Quint).

The Ministry of Health and Family Welfare launched JANANI (“Journey of Antenatal, Natal and Neonatal Integrated Care”), a national digital tracking platform for pregnancies, childbirth, newborn care, and women’s reproductive health, in May 2026. Registration mechanisms include Aadhaar number, ABHA (Ayushman Bharat Health Account) ID, Aadhaar-based OTP, biometric verification, or mobile number; the default path is Aadhaar (Business Today, 10 May 2026).

A 2018 replication study of the Janani Suraksha Yojana (JSY) scheme (Lim and Eckholm) documented 24.3% of non-recipients citing bank-account absence and 9.7% citing missing Aadhaar number as primary reasons for non-receipt of JSY institutional-delivery cash benefits (Journal of Development Studies). A 2022 study, “A decade of conditional cash transfer programs for reproductive health in India: How did equality fare?”, found the most-marginalized populations (Scheduled Tribes, illiterate women, lowest-wealth quintile) bear disproportionately the burden of Aadhaar-linkage failures (PMC).

4.2 Worldcoin / World (iris-biometric identity)

Worldcoin (Tools for Humanity), associated with Sam Altman and Alex Blania, launched in July 2023 with the “Orb” iris scanner producing an IrisCode-keyed “World ID” credential (Wikipedia: Worldcoin).

Spain’s Agencia Española de Protección de Datos (AEPD) became the first national regulator to ban Worldcoin operations (18 March 2024), citing insufficient information at orb-scan intake, scans of minors, and inability to withdraw consent; it ordered deletion of all Spanish-user biometric data in December 2024. The stated grounds are the regulator’s findings (Biometric Update, March 2024).

Hong Kong’s Office of the Privacy Commissioner for Personal Data ordered Worldcoin to cease operations (22 May 2024), ruling that the planned ten-year retention of iris biometric data for AI training was unjustified and that operators had not verified scan subjects were over eighteen (Biometric Update, May 2024; DL News).

Worldcoin paused Orb verification in India, Brazil, and France in December 2023 (TechCrunch).

Match Group announced a pilot integration of World ID iris-scan verification on Tinder (2026), beginning in Japan with a planned US rollout — the first dating-platform integration, framed as “verified-human” trust signaling (Rest of World, March 2026). Reproductive-stack significance: dating-app verification is the precursor data layer to fertility-app and IVF-platform identity binding.

Worldcoin announced 10 million verified users despite operating bans in seven jurisdictions — the company’s own announcement (Bitget News).

4.3 US state reproductive-identity and surveillance laws

Texas SB 1188 (signed 2025, effective 1 September 2025) requires Texas electronic health records to include an unalterable field for “biological sex at birth,” defined as “the reproductive trait determining production of male or female gametes.” Changes are permitted only for clerical errors or diagnosed sexual-development disorders (intersex conditions). It grants parents and guardians unrestricted access to minors’ EHRs, sets a maximum penalty of $250,000 per noncompliance instance, and its EHR-storage provisions took effect 1 January 2026 (Texas Tribune, 22 May 2025; LegiScan enrolled text).

Texas HB 229 (signed 2025) defines “woman” in state law as “an individual whose biological reproductive system is developed to produce ova,” and “man” as “an individual whose biological reproductive system is developed to fertilize the ova of a female,” applying that definition across statutes (Wikipedia: Texas HB 229).

Texas SB 2880 (signed August 2025, effective 4 December 2025) allows any private citizen to sue manufacturers, distributors, or mailers of abortion medication with a $100,000 minimum statutory damages award; Governor Greg Abbott’s signature followed Senate passage on 1 May 2025 (Texas Tribune, 22 August 2025; KUT).

Texas SB 31, the Life of the Mother Act (2025), clarifies when emergency abortion is legal and adds mandatory training requirements for physicians and lawyers, with bipartisan Senate support (Equality Texas summary).

Idaho’s “abortion trafficking” statute (signed April 2023) criminalizes obtaining abortion medication for, or transporting/harboring, an unemancipated minor for an out-of-state abortion without parental consent, with a penalty of 2–5 years. U.S. Magistrate Judge Debora Grasham preliminarily enjoined the law in November 2023. The Ninth Circuit reversed in part on 2 December 2024, allowing enforcement of the “harboring” and “transporting” provisions while finding the “recruiting” language overbroad and unconstitutional under the First Amendment (Idaho Capital Sun; First Amendment Center; CBS News signing report).

Texas AG Ken Paxton’s expanding litigation (Texas v. Aid Access et al., S.D. Tex., filed 2026) targets out-of-state physicians alleged to mail abortion medication into Texas; the first individual complaint is positioned for U.S. Supreme Court review (Houston Public Media, 25 February 2026; Axios Houston, 4 May 2026).

Half a dozen state attorneys general issued post-Dobbs (June 2022) consumer alerts warning of menstrual-tracking app data exposure. A subsequent investigation by the Leadership Center for Attorney General Studies documented sixteen apps with explicit subpoena-disclosure clauses. The Nebraska Facebook-message subpoena case (2022) established the precedent that messaging data — not period-tracker data per se — has been the primary source of digital evidence in post-Dobbs prosecutions (Leadership Center for AG Studies; Stateline).

5. AI in IVF clinics — embryo grading and viability prediction

The predictive-performance claims for each scoring product below are the vendor’s own product description. Independent RCT-class evidence (the Nature Medicine 2024 trial) reports noninferiority to expert morphology, not superiority — the vendor claims are not upgraded on the strength of it.

5.1 Time-lapse imaging and morphokinetic AI

Vitrolife’s iDAScore is a deep-learning model trained on time-lapse image sequences from EmbryoScope incubators across the first six days of embryo development; per Vitrolife it returns a probability score that a transferred embryo will implant and progress to clinical pregnancy. It holds CE-mark certification in Europe and is not FDA cleared in the US (marketed there as “investigational” only) (Vitrolife product page; PMC method paper, “Robust and generalizable embryo selection based on artificial intelligence and time-lapse image sequences”).

Life Whisperer (Presagen, Australia) markets a two-product line — Life Whisperer Viability (a 0–10 score for clinical-pregnancy likelihood) and Life Whisperer Genetics (a 0–10 score for euploidy likelihood) — trained on static microscopy images rather than time-lapse video. It is available in international markets and not sold in the United States, indicating no FDA marketing authorization (Life Whisperer site; PMC validation paper).

5.2 Multi-modal and transformer-based models

The IVFormer / VTCLR (Vision Transformer Contrastive Learning Representation) self-supervised pipeline (2024–2025) covers pronuclear-stage symmetry, cleavage-stage blastomere features, and blastocyst-stage inner-cell-mass and trophectoderm grading in a single model (ScienceDirect).

Yanran Yang et al., “Deep learning versus manual morphology-based embryo selection in IVF: a randomized, double-blind noninferiority trial” (Nature Medicine, June 2024), is the first published RCT-class evidence on AI embryo selection in clinical IVF, reporting noninferiority of the model versus expert embryologist morphology assessment on cumulative live-birth rate (Nature Medicine).

“Deep learning classification integrating embryo images with associated clinical information from ART cycles” (Scientific Reports, 2025) performs multimodal fusion of images plus patient-clinical variables such as age, AMH, and BMI (Scientific Reports). A 2025 scoping review in Frontiers in Reproductive Health covers deep-learning applications for human embryo assessment via time-lapse imaging from 2020 through 2024 (Frontiers in Reproductive Health).

5.3 Pregnancy-outcome prediction and meta-evidence

“Predicting pregnancy outcomes in IVF cycles: a systematic review and diagnostic meta-analysis of artificial intelligence in embryo assessment” (2025) pools diagnostic-accuracy metrics across studies (PMC). “A multiday machine learning framework based on improved-NEQsi for predicting embryo quality and pregnancy outcomes in IVF” appeared in the Journal of Assisted Reproduction and Genetics (Springer, 2026) (Springer). “Artificial Intelligence-Empowered Embryo Selection for IVF Applications: A Methodological Review” appeared in MDPI Machine Learning and Knowledge Extraction (2025) (MDPI).

5.4 Regulatory posture

As of May 2026, no fertility-clinic AI embryo-selection product has received FDA marketing authorization (510(k) clearance or De Novo) for use as a diagnostic device in the US. US clinics that deploy time-lapse AI scoring either treat the score as decision support for the embryologist or operate under the Clinical Laboratory Improvement Amendments (CLIA) lab-developed-test pathway; the European CE mark has been the primary regulatory route. The regulatory gap is consequential because the upstream pipeline — Orchid, Heliospect, Herasight, Nucleus polygenic scoring — runs before the AI selection layer. A complete future embryo-selection workflow chains polygenic scoring → time-lapse AI scoring → transfer decision, with no FDA review of either layer in the US clinical market (Vitrolife product page; Life Whisperer site; Journal of IVF-Worldwide review).

6. “Good genes” rhetoric — political and Silicon Valley

The quoted words below are verbatim on-record and preserved as spoken. Every “eugenics” / “consumer eugenics” / “racial eugenics” characterization is the named outlet’s or scholar’s, attributed — not this page’s voice, and not sanded into anodyne paraphrase. The page adopts neither the critics’ pejorative nor the industry’s “empowering choice”; the words are quoted and the framing is left to the reader.

6.1 Political statements on heredity

At a Bemidji, Minnesota rally (18 September 2020), Donald Trump said: “You have good genes. A lot of it is about the genes, isn’t it, don’t you believe? The racehorse theory.” The quotation is on the record; the “racial eugenics” framing in the headline is The Hill opinion writer’s characterization (The Hill).

On Hugh Hewitt’s program (October 2024), Trump stated the United States had “a lot of bad genes in our country right now” in the context of immigration (STAT, Eric Boodman, 28 October 2024).

Trump’s “poisoning the blood” line (December 2023 and subsequent rallies) was repeated in campaign speeches; the Hastings Center for Bioethics linked the construction to early-twentieth-century immigration-restriction eugenics — the phrase is on the record, and the eugenics-genealogy linkage is the Hastings Center’s attributed analysis (Hastings Center).

Shannon Bow O’Brien (UT-Austin presidential-speech scholar) published a 2024 book-length argument that Trump’s rhetoric draws explicitly on American eugenic-tradition vocabulary (Center for Genetics and Society interview). Scientific American ran an October 2024 editorial, “Donald Trump Wants to Make Eugenics Great Again. Let’s Not.” (Scientific American).

6.2 Silicon Valley pronatalist circuit

The Natal Conference (NatalCon), founded by Kevin Dolan, was held in Austin in 2023 with a second iteration in 2025. Speakers included Malcolm and Simone Collins, who run the Pronatalist Foundation; the first conference featured a presentation by Charles Haywood (Foundationalist Press) on men-only spaces, and the 2025 Austin event included a Collins-hosted cocktail party with founders of gene-editing and genetic-engineering companies present (NPR, 25 April 2025; Slate; Washington Post profile, 1 February 2025; Natalism.org event page).

The Collinses have publicly stated they used Genomic Prediction’s PGT-P for their own children — their own on-record statements, as reported by Slate (Slate profile above; Wikipedia: Simone and Malcolm Collins).

JD Vance said “I want more babies” at the March for Life (January 2025), in his capacity as Vice President (KFF Health News). The White House executive order “Expanding Access to In Vitro Fertilization” (signed 18 February 2025) did not change existing IVF policy but framed “family formation” as a national priority (NPR). EMD Serono (Merck KGaA) agreed with the administration in October 2025 to lower the price of its fertility-drug product line as part of the IVF-access initiative (KFF Health News). Transportation Secretary Sean Duffy directed his department (2025 memo) to prioritize federal funding for jurisdictions with high marriage and birth rates (KFF Health News).

6.3 Silicon Valley philanthropic and investment ties

Brian Armstrong (Coinbase CEO) is an investor in Orchid Health and separately announced a $30M investment via the new venture Preventive into embryo gene-editing (CRISPR-class) work (MIT Technology Review, Antonio Regalado, 5 June 2025; Center for Genetics and Society).

Marc Andreessen’s “Techno-Optimist Manifesto” (16 October 2023), ~5,200 words in ~108 numbered statements across fifteen parts, cites Nick Land as a “patron saint” and names F.T. Marinetti’s 1909 Italian Futurist Manifesto as a structural influence (a16z; Wikipedia: Techno-Optimist Manifesto; Jacobin critique).

Elon Musk has made repeated public statements on low birth rates as “civilizational risk,” with public-record acknowledgment of multiple children by multiple partners (including via IVF), and has endorsed large family sizes as the dominant near-term challenge; coverage in NPR, CNN, and the Washington Post profiles above places Musk as a rhetorical ally of the NatalCon / Collins circle (CNN, 10 April 2025).

6.4 Critical framing voices

Right-of-center outlets have applied the “consumer eugenics” label as well as left-of-center ones — a fact that complicates the easy left-versus-right read of the debate, and one this page records without adopting either side’s framing. Christine Rosen wrote “Silicon Valley’s Consumer Eugenics” for Commentary Magazine (Commentary); The Heritage Foundation published “The Rise of Consumer Eugenics” (Heritage); and the AEI Center for Technology, Science, and Energy ran “Silicon Valley’s Consumer Eugenics” (AEI CTSE).

The Center for Genetics and Society (Marcy Darnovsky, executive director) is a long-running progressive bioethics watchdog with systematic coverage of polygenic-screening startups, including a named-company timeline back to 2017 (Center for Genetics and Society). Hank Greely (Stanford Law) and Rob Klitzman (Columbia) published “Genetic testing of embryos: An ethical debate unfolds” (Boston Globe, 7 December 2023) (Boston Globe). Ben Williamson (Code Acts in Education) published “Genetic IQ Tests Are Bad Science and Big Business” (July 2024), a critique of the educational-genomics pipeline that feeds the embryo-IQ pitch (NEPC).

Through-line

The six threads connect in a single stack:

  1. Identity binding (Aadhaar, Worldcoin, Texas SB 1188). A child’s sex is fixed at birth in an unalterable EHR field; an adult’s reproductive-benefit eligibility is gated by biometric authentication; a dating-app match is gated by an iris scan. The reproductive trajectory becomes legible to the state and to commercial intermediaries from before conception.

  2. Genetic-database accumulation (23andMe → TTAM, UK Biobank, Heliospect, the Henrietta Lacks pattern, Florida HB 833). The genetic substrate is owned, sold, and litigated. The “exclusive property” language of Fla. Stat. § 760.40 is the property-law half of the bargain; the bankruptcy-asset disposition of 23andMe to TTAM is the contradiction. The user owns the property until a Chapter 11 court sells it.

  3. Polygenic scoring pipeline (Orchid, Genomic Prediction, Heliospect, Nucleus, Herasight). Whole-genome sequencing of embryos plus polygenic-risk scoring derived from population-scale biobanks — into a regulatory void: no FDA, no NICE, no ACMG-clinical authorization for the trait-prediction use case; statutory prohibition in the UK but not the US.

  4. Clinical AI selection (iDAScore, Life Whisperer, IVFormer). The embryo that survives polygenic ranking is ranked again by morphokinetic AI before transfer. Two AI layers, neither FDA-cleared in the US clinical market.

  5. Algorithmic family policing (AFST and its replicas). After the child is born, the same statistical-prediction logic is applied to whether the parent gets to keep her. Contact with public services is a feature; disability is a feature. The Hackney case is the named-plaintiff exemplar of the post-natal arm.

  6. Rhetoric (Trump, Vance, Andreessen, Musk, the Collinses). The political surround names the project openly: “good genes,” “bad genes,” “family formation,” pronatalism, the Techno-Optimist Manifesto’s nineteenth-century vocabulary. The startups are the instrument; the rhetoric is the demand-generation.

A reproductive trajectory that was once a private matter between adults is now a routed transaction across a vendor stack. Each layer is independently legal, independently funded, independently defensible on its own terms. The thing to notice is the chain. No single vendor is “eugenic.” The chain selects.

Verdict

The evidentiary spine here is documentary and defensible: the statutes (Florida HB 833, the Texas suite, Idaho’s trafficking law), the court dockets and settlements (Lacks, 23andMe → TTAM, Natera, the Hackney case cleared to proceed), the regulator bans (Worldcoin in Spain and Hong Kong), the peer-reviewed IVF-AI literature (the Nature Medicine 2024 noninferiority RCT), and the company foundings, funding rounds, named backers, and published prices.

The line that must not be crossed is the efficacy line. Every “select embryos, gain N IQ points” figure (Hsu’s ~15, Heliospect’s ~6, Herasight’s cognitive predictor) and every disease-risk-reduction percentage (Herasight’s 20–44%) is a company or advocate claim that peer review does not support: ACMG called clinical use “premature” (2023), Karavani et al. (eLife 2022) reached the same conclusion, and named scientists (Pachter, Gusev) contest the within-family predictive validity directly. The record can report that the companies sell this promise; it cannot report that the promise is delivered.

On the politics: the actual words (Trump, Vance, Andreessen, Musk, the Collinses) are verbatim on-record, and the “eugenics” / “consumer eugenics” framing belongs to the outlets and scholars who apply it — The Hill, STAT, the Hastings Center, Scientific American, Bow O’Brien, the Center for Genetics and Society, Slate, and, notably, the right-of-center critics (Commentary, Heritage, AEI) whose parallel objection complicates the easy left-versus-right read. Neither the critics’ pejorative nor the industry’s “empowering choice” is adopted here. The behavior is on the record; the framing is the reader’s to weigh.

Get updates on the Evil Robots series

Newsletter essays on AI escape, deception, and the humans who built them.