health
Heritable human genetic enhancement beyond disease correction occurs in at least one jurisdiction by 2126.
Last revised Sep 13, 2026.
Current estimate
probability this has happened by each date
The full criterion above is met by 31 December 2035. New this cycle; this subject had no horizon grid before. Near the floor, and from a count rather than a judgment: eight years after the only case there has been no second one, no funded entity has named a target locus, the one person who names a target cannot leave China and says he has not begun embryo work, and the best available embryo-editing data reports off-target editing at eleven of fourteen candidate sites with mosaicism in most one-cell injections. Nine years is also shorter than the thirteen-month verification lag plus the gestation and the pre-clinical work, so even a decision taken today resolves late in this window.
The full criterion above is met by 31 December 2050. Calibrated against the nearest reproductive-technology precedents rather than against sentiment. Mitochondrial replacement took nine years from United Kingdom legalisation to eight licensed births, and its unregulated route produced its first birth within a year of the United States bar; so a determined actor with a venue is a decade-scale story, not a five-year one. The counterweight is the cloning null: thirty years of demonstrated mammalian capability under a comparable procedural bar has produced zero human instances.
A verified live birth of a human carrying an intentional heritable modification where, at the modified locus, the installed genotype confers an advantage relative to the population-typical genotype of that embryo's ancestry-matched reference population rather than restoring it to that genotype, and where human evidence of that advantage for that specific allele existed before the edit was made. Assessment is from the genome and the published human literature, not from the operator's stated purpose and not from any documented indication. Instrument: none exists. The WHO Human Genome Editing Registry defers human embryos to a second phase that has not begun, no national assisted-reproduction registry was found to record embryo genetic modification, and the only case ever resolved was resolved by a criminal court roughly thirteen months after the births. Resolving power for a covert case is zero.
Resolution criteria
- 2035
- The full criterion above is met by 31 December 2035. New this cycle; this subject had no horizon grid before. Near the floor, and from a count rather than a judgment: eight years after the only case there has been no second one, no funded entity has named a target locus, the one person who names a target cannot leave China and says he has not begun embryo work, and the best available embryo-editing data reports off-target editing at eleven of fourteen candidate sites with mosaicism in most one-cell injections. Nine years is also shorter than the thirteen-month verification lag plus the gestation and the pre-clinical work, so even a decision taken today resolves late in this window.
- 2050
- The full criterion above is met by 31 December 2050. Calibrated against the nearest reproductive-technology precedents rather than against sentiment. Mitochondrial replacement took nine years from United Kingdom legalisation to eight licensed births, and its unregulated route produced its first birth within a year of the United States bar; so a determined actor with a venue is a decade-scale story, not a five-year one. The counterweight is the cloning null: thirty years of demonstrated mammalian capability under a comparable procedural bar has produced zero human instances.
- 2126
- A verified live birth of a human carrying an intentional heritable modification where, at the modified locus, the installed genotype confers an advantage relative to the population-typical genotype of that embryo's ancestry-matched reference population rather than restoring it to that genotype, and where human evidence of that advantage for that specific allele existed before the edit was made. Assessment is from the genome and the published human literature, not from the operator's stated purpose and not from any documented indication. Instrument: none exists. The WHO Human Genome Editing Registry defers human embryos to a second phase that has not begun, no national assisted-reproduction registry was found to record embryo genetic modification, and the only case ever resolved was resolved by a criminal court roughly thirteen months after the births. Resolving power for a covert case is zero.
Probability history
| Revised | State | p(2035) | p(2050) | p(2100) |
|---|---|---|---|---|
hf-forecast-0211 | researched | 2% no prior revision | 12% no prior revision | — |
hf-forecast-0204 | researched | — | — | — |
hf-forecast-0024 | researched | — | — | — |
hf-forecast-0007 | initial prior | — | — | — |
Strongest evidence for
- The act has already happened once. Three children were born with intentional heritable edits within about six years of CRISPR becoming usable, carried out by one funded individual against universal condemnation, and the fact was established by a criminal court (hf-evidence-0200).
- A maintained catalogue of more than eighty characterised protective human variants exists, which is exactly the list the repaired criterion's prior-human-evidence clause requires an actor to pick from (carried forward from hf-evidence-0146).
- The one person who has actually done this states an intention to install a naturally occurring Alzheimer's-protective variant found in Icelandic and Scandinavian people - a qualifying allele under the repaired criterion, named by the only actor with a demonstrated willingness to proceed (hf-evidence-0199).
- A reproductive technique barred in the United States only procedurally did relocate offshore and produced double-digit births within three years, while the regulated route took a decade to produce eight (hf-evidence-0201).
- Prospera ZEDE's own published rules are genuinely silent on embryo genetic modification, confirmed against the zone's own text rather than press characterisation (hf-evidence-0202).
- Base editing in human embryos produced no segmental aneuploidies or large deletions at the target sites, removing the chromosomal-damage objection that could not be engineered around (hf-evidence-0197).
Strongest evidence against
- The support cycle 18 leaned on hardest is withdrawn by its own source: the embryo base-editing study says its targets were chosen for zygote availability and validated guides, 'not therapeutic relevance to germline gene editing', and that neither edit would be eligible for germline use under National Academies guidance (hf-evidence-0197).
- The first pre-registered discriminator returned empty. No funded entity has named a first target locus; every one that says anything commits to severe genetic disease (hf-evidence-0199).
- The second pre-registered discriminator fired with the wrong sign. No jurisdiction moved from silence to permission since 2024, and Japan moved from unpenalised guidance to up to ten years' imprisonment (hf-evidence-0198).
- Human reproductive cloning is the closest structural analogue to the seed's own relocation argument and refutes it: thirty years, demonstrated across roughly seven mammalian species, a procedural United States bar, and zero human instances (hf-evidence-0201).
- The safety picture is two-sided, not clean: base editing at eleven of fourteen candidate off-targets, mosaicism in most one-cell injections, and the senior author's own summary that 'it is currently not possible to do so safely' (hf-evidence-0197).
- No instrument exists that could score this forecast. The WHO registry's embryo phase never began and no assisted-reproduction registry records genetic modification, so resolution depends on publicity, investigation or prosecution (hf-evidence-0200).
- South Africa never held an explicit statutory permission; what was repealed in 2025 was ethics-council guidance, and the underlying statute bars genetic manipulation only for reproductive cloning and was ambiguous throughout (hf-evidence-0202).
Present uncertainties
- Whether an organised constituency for heritable enhancement as distinct from heritable disease prevention ever forms, which the revised structural argument makes the decisive variable and which no evidence in this cycle measures. Whether mosaicism and off-target editing are tractable at all in embryos, where the first good data is two-sided. Whether the prior-human-evidence clause will still be the right test in a century, given that it gets easier to satisfy as the catalogue grows.
Next discriminating observation
First, and chosen because the revised structural argument makes it decisive rather than merely interesting: any organised demand-side signal for enhancement specifically - a patient or consumer group, a waiting list, a priced offer, a clinic advertising a non-disease heritable trait - as opposed to the supply-side company announcements this seed has been counting. The cloning null says supply without demand produces nothing for thirty years, and every actor this seed has catalogued is supply. Second: whether He Jiankui's stated Alzheimer's-protective-variant plan produces any concrete step - an embryo, an institutional affiliation, a named gene rather than a described phenotype - since under the repaired criterion he is the only actor whose stated target would qualify.
Sources
- The PCSK9 embryo study says in its own words that its target was chosen for convenience and is not eligible for germline editing, which withdraws the directional signal this seed drew from it in cycle 18
Cycle 18 recorded, in hf-evidence-0145 and in the factor-two reasoning of hf-forecast-0204, that 'the first human-embryo base-editing study chose PCSK9, a protective locus, not a disease correction', and treated that as evidence about where the field is heading. The preprint's own full text says otherwise: 'the choice of editing targets ... was driven by the availability of normal human zygotes and previously tested sgRNAs, not therapeutic relevance to germline gene editing.' It adds: 'Neither edit introduced here would be eligible for germline gene editing according to the latest report by the National Academies of Sciences, which recommends that the introduction of only naturally occurring and common genetic variants should be considered for heritable gene editing.' The edit itself is not the natural protective variant but an engineered one - an 'ABE-mediated A-to-G edit at the splice site between exon 1 and intron 1 of the PCSK9 gene, resulting in read-through and premature termination' - and the second target was an 'A-to-G substitution at the -198 position in the promoter region of the HBG1 and HBG2 genes'. The study's safety result is two-sided rather than clean: no segmental aneuploidies or large deletions at the target sites, but 'base editing at 11 out of the same 14 candidate off-targets', and 'in most cases, injection at the 1-cell stage results in mosaic embryos'. The senior author's summary in the institutional release is 'it is currently not possible to do so safely'.
Limit: This is one study by one group and it bears on the direction of a research field, not on what a private actor would choose. It does not show that nobody will install a protective allele; it shows that this particular case cannot be cited as evidence that someone has, which is the narrower thing the ledger was doing with it.
The bioRxiv full text was fetched directly by the master agent and every quoted sentence above was read in place; the Columbia release was also fetched directly. The Nature version of record is paywalled and was not fetched. A delegated lane had reported the same paper as a protective-variant installation and could not render the primary page; the correction came from reading it.
- The one jurisdiction that moved since 2024 moved toward criminalisation, not permission: Japan replaced unpenalised guidance with up to ten years' imprisonment
Japan enacted the Act on Regulation of Handling of Human Genome-Edited Embryos on 17 July 2026, promulgated as Act No. 70 of 2026. Reported text: 'The bill prohibits the transplantation of embryos created by genome-editing sperm or eggs, as well as genome-edited embryos, into human or animal uteri', and 'The bill sets out penalties of up to 10 years of imprisonment or a fine of up to 10 million yen, or both, for any violations.' Before this, Japan's 2019 position was non-binding guidance that permitted basic research and disallowed uterine transfer while carrying no penalty. The Nature Medicine commentary on it is titled 'A binding ban, and its limits'.
Limit: The prohibition attaches to transfer into a uterus, not to creating or researching an edited embryo, so it does not stop the upstream work; the Nature Medicine title's word 'limits' probably refers to exactly that, though the article was not read. Japan was never one of the venues this forecast's high estimates depended on, so the direct effect on the probability is small.
The Star's copy was fetched directly by the master agent and both quoted sentences were read in place. Three further attempts to reach primary or near-primary copies failed - Japan Times returned HTTP 402, Asia News Network 403, and the Science Japan item 404 - so the statute's own Japanese text was never opened. The prohibition's scope and penalty are corroborated across three independent outlets including the Japan Science and Technology Agency's own service, which is why they are recorded; the Act number and enactment date rest on fewer sources and are the weaker half of this record.
- The first pre-registered discriminator returns a clean negative: no funded entity has named a target locus, and the only person naming a target names a phenotype and is confined
Cycle 18 pre-registered this observation: if a funded entity states that its first edit is at a locus where the embryo carries no disease variant, factor two resolves most of the way upward. A directed search across the known actors returned no such statement. Preventive's own announcement commits to 'preventing severe genetic diseases' and 'the underlying genetic cause' and names no gene; a separate report on the company found no locus either, and its chief executive denied the specific-couple discussions reported elsewhere. Nucleus Genomics' chief executive describes editing only as a long-term goal - 'If we can show both efficacy and safety, eventually, it's a long-term goal, but the goal would be to bring gene editing into clinical practice' - with no gene named, and its shipped product is embryo screening. Manhattan Genomics' first target could not be confirmed by any successful fetch. The only actor naming a target is He Jiankui, who describes an intention to 'modify one letter in the human DNA sequence to simulate a natural mutation found in some Icelandic and Scandinavian people', with the personal motive 'I decided to do Alzheimer's disease because my mother has Alzheimer's'; he names a phenotype and a population rather than a gene symbol, is unaffiliated and funded by private donations in Hainan, and the same reporting places him under travel restriction.
Limit: Absence of a public statement is weak evidence about private intent, and it is the weakest kind of negative: a company with a target has every commercial and legal reason not to announce it. This record shows the discriminator did not fire, not that no such target exists.
The MIT Technology Review interview, preventive.bio's announcement and the Futurism report were each fetched directly by the delegated lane, which reported fetch status per source; the master agent did not re-fetch them, and the He Jiankui interview is prior-cycle vintage re-confirmed rather than new. The gene symbol APP A673T is not stated in the interview and is an inference from the described phenotype, recorded here as an inference and not as a quote. A search-derived claim that He Jiankui is currently affiliated with a sequencing company was checked, found to be stale, and is not recorded.
- There is no instrument: the one registry built for this question defers embryos to a second phase that has not arrived, and the only case ever resolved was resolved by a criminal court thirteen months late
The WHO Human Genome Editing Registry page states that 'In a first phase, the HGE Registry will cover both somatic and germline clinical trials', and that 'In a second phase, the HGE Registry will also cover research using genome editing technologies on (i) human embryos and on (ii) germline cells or their progenitors when gametes derived from these will be used to create embryos.' The page carries no entry count, no last-updated date and no indication the second phase has begun. No registered heritable-editing study with intent to establish a pregnancy was found on ClinicalTrials.gov or the WHO ICTRP; the trials catalogued there are somatic. No national assisted-reproduction outcome registry - the United Kingdom's HFEA, the United States SART series, or the ESHRE European register - was found to contain a field recording whether an embryo was genetically modified, though this could not be confirmed against their data dictionaries. The instrument that actually resolved the only real case was a criminal prosecution: the Nanshan District People's Court of Shenzhen sentenced He Jiankui to three years' imprisonment and a three million yuan fine 'for illegally carrying out human embryo gene-editing intended for reproduction', and stated that 'three genetically edited babies were born'. The births were announced in November 2018 and the verdict is dated 30 December 2019, a lag of roughly thirteen months.
Limit: A registry's absence today says little about a hundred-year horizon; registries get built. The finding bears on resolving power now and on the shape of any near-term resolution, not on whether the event occurs.
The WHO registry page was fetched directly by the master agent and the two phase sentences were read in place; the Xinhua verdict article was fetched directly by a delegated lane. The registry absences are negatives established by search rather than by querying the databases, and the assisted-reproduction registry question was not settled either way - none was checked against its own data dictionary, and that is recorded as unestablished rather than as a null.
- Two base rates that point opposite ways: mitochondrial replacement relocated offshore and produced births, and human reproductive cloning produced none in thirty years under the same kind of bar
The FDA's own advisory states that 'Since December 2015, Congress has included provisions in annual federal appropriations laws that prohibit FDA from accepting applications for clinical research using MRT', and that 'FDA maintains the authority to investigate and take enforcement action in the event that it becomes aware of noncompliance with the laws and regulations administered by FDA.' Under that bar the technique did relocate: a first birth in Mexico in 2016 from a United States team, then a Kyiv clinic reporting seven births by December 2018, with one practitioner later self-reporting a personal total of fifteen children. The regulated route produced far fewer and far later: the United Kingdom legalised the technique in 2015 and Newcastle reported eight babies born to seven women in July 2025, a decade later. The countervailing base rate is reproductive cloning. Dolly was born in July 1996, the technique diffused across roughly seven mammalian species, the United States bar is procedural rather than criminal, and thirty years on no human reproductive clone has ever been born. The only United States enforcement ever exercised in the mitochondrial case was a warning letter in August 2017 to the team that had created the embryo on United States soil before exporting it, which locates the legal hook at the place of manufacture rather than at the place of implantation.
Limit: Neither analogue is clean. Mitochondrial replacement had an organised patient constituency with a specific disease, which heritable enhancement does not obviously have, and the jurisdictions that hosted it were permissive by absence rather than by design. Cloning is confounded by near-universal criminal bans elsewhere and by a much higher technical failure rate, so its null may measure the technique rather than the instrument.
The FDA advisory page was fetched directly by a delegated lane and both quoted sentences come from it. The Newcastle figures, the Kyiv and Mexico birth counts, the fifteen-child self-report, the Dolly figures and the cloning null are search-result synthesis of the cited pages rather than fetched primary text, and the fifteen-child figure is self-reported by the practitioner and was not independently audited. The primary text of the 2017 warning letter was not obtained; its contents are secondary throughout.
- The permissive-venue search returns one verified silence and two corrections to press framing, including a second instance of the same error this seed already caught once
Prospera ZEDE's own published rules corpus contains no mention of embryo genetic modification, germline editing or assisted-reproduction regulation of any kind, confirmed by direct fetch of the ZEDE's own page; this is a verified silence at the zone level rather than an inferred one, though Honduran national law could not be obtained and the claim that gene-therapy clinics there operate 'without regulation by the Honduran government' remains a secondary characterisation of an absence. South Africa's position is not what the press framing says. The National Health Act 2003 section 57(1) prohibits genetic manipulation only 'for the purpose of the reproductive cloning of a human being', and section 57(4) provides that 'The Minister may permit research on stem cells and zygotes which are not more than 14 days old on a written application'. What was adopted in 2024 and repealed in 2025 was National Health Research Ethics Council guidance, not statute, and the statute itself was ambiguous rather than an explicit permission. The Emirates' Federal Law No. 7 of 2019, Article 14, as amended by Federal Decree-Law No. 17 of 2023, prohibits centres from 'making non-therapeutic genetic modifications' to fertilised or unfertilised eggs and sperm 'even if upon the consent of the spouses'. No jurisdiction anywhere was found to have moved from silence to explicit permission since 2024.
Limit: A verified silence in a special economic zone is not a permission, and a zone that is silent on embryo editing is also silent on everything that would make it practicable. Nothing here establishes that any venue would accept the work, only that one venue has not written it down.
Only two of these rest on fetched text. The Prospera page and the South African article, including both quoted sections of the National Health Act, were fetched directly by a delegated lane. The Emirates article was not: the government legislation portal returned 403 and its mirror refused the connection, so the Article 14 wording is search-result synthesis of the portal's indexed text and is recorded at that tier deliberately. The named list of countries with no relevant policy was never obtained, so the seed's long-standing ten-country figure remains unverified against any national statute, as hf-evidence-0147 already warned.