Biological Enhancement

Biological enhancement is intervention in a person who is not ill, intended to push some capability past a normal baseline rather than to restore something lost. The framing under test is that enhancement is a coherent category, cleanly separable from therapy. The most useful available answer is that the body which made the distinction famous examined it and concluded it does not hold — which is a far more interesting starting point than the usual gloss, and it is on the record.

Established The measured content of this subject is smaller and stranger than its literature. Established The best-characterised human enhancement genotype is a single child with a myostatin splice mutation whose quadriceps measured 7.2 standard deviations above matched controls. Established The best-quantified enhancement of a healthy adult is a shoe-mounted machine that reduces the metabolic cost of walking by about a sixth. Established The most reliable finding in cognitive enhancement is a negative one: training produces enormous gains on the trained task and essentially none anywhere else. Frontier And the enhancement technology that is actually being sold today is not an edit or a drug at all — it is polygenic scoring of embryos, marketed with a number that does not mean what buyers take it to mean.

Speculative Underneath the commerce sits a genuinely open question that this brief takes seriously: whether human cognition has real headroom, or whether the ceiling reached by ordinary variation is close to the ceiling that any intervention can reach. Speculative The strongest arguments that there is no headroom are quantitative and recent, and they come from within the field. Handwave The strongest arguments that there is headroom require a mechanism nobody has demonstrated: not a trained skill, not a shifted allele frequency, but something that changes the developmental rules. Frontier A reader should leave with the shape of the constraint rather than a verdict: the binding link in this chain is not a laboratory result, it is a decision no regulator has been asked to make, because every approved intervention in this space carries a disease indication and there is no pathway for one that does not.

Established The therapy/enhancement boundary was drawn most carefully by the people who then declared it unusable. Beyond Therapy: Biotechnology and the Pursuit of Happiness, the 2003 report of the US President’s Council on Bioethics under Leon R. Kass, is organised into four chapters — Better Children, Superior Performance, Ageless Bodies, Happy Souls — and defines therapy as restoring a person to a normal state of health and enhancement as altering the normal workings of the body and psyche to augment native capacities. Established It then concludes that the distinction is “finally inadequate to the moral analysis,” because health and normality are themselves contested and the categories overlap in practice. Established The report’s own worked example is growth hormone for short stature: the same molecule, the same centimetres, sorted into therapy or enhancement by whether a diagnosis happens to be available. Frontier That is not a philosophical quibble. It is the reason the regulatory position described in section 9 exists, and it predicts the failure mode of every attempt since to police enhancement by defining it.

Established The cleanest natural human enhancement phenotype is one child, and it is a loss-of-function mutation. Schuelke et al. reported in the New England Journal of Medicine in 2004 a boy with a g-to-a transition at the myostatin gene’s first intron splice donor, causing aberrant splicing; he was homozygous, his mother heterozygous. Established His quadriceps cross-sectional area was 6.72 cm² against 3.13 ± 0.49 cm² in ten age- and sex-matched controls — 7.2 standard deviations above the control mean — with subcutaneous fat 2.88 standard deviations below, and no mature myostatin detectable in serum. Established The mother, a heterozygote and a former professional athlete, was muscular and otherwise unremarkable. Frontier Two things follow that enhancement writing usually drops. The phenotype required complete loss of a functioning pathway, so the dose-response is not gentle; and a single child is a case report, not a safety dataset, which is why myostatin and activin pathway inhibitors have repeatedly added lean mass in adults without a proportional gain in function. Frontier Specific trial figures for those drug programmes could not be verified for this brief and none are printed here.

Established Genetic performance enhancement was banned before it existed, and twenty-three years later there is still nothing to detect. The World Anti-Doping Agency added gene doping to the prohibited list in 2003, defining it as any non-therapeutic genetic manipulation for enhancement of athletic performance, with a named candidate list: EPO, IGF1, MSTN, PPARG — the “marathon mice” with roughly double the endurance of untreated animals — ACTN3, and VEGF, which was tested and gave poor long-term results. Established There is exactly one documented case in the record and it is an attempted purchase: in 2006 a German coach was convicted of administering performance-enhancing drugs without athlete knowledge, and the gene-doping element was an email read into open court in which he tried to obtain Repoxygen, an erythropoietin gene-therapy construct. Established No confirmed instance of gene doping in competition has ever been detected. Frontier Detection is indirect and structurally hard: the main approaches are PCR against exon–exon junctions in complementary DNA, absent from genomic DNA, and next-generation sequencing — and natural variation mimics the target, the canonical case being the Olympic skier Eero Mäntyranta, whose elevated red-cell mass came from an erythropoietin-receptor variant. Speculative The most economical reading of a twenty-three-year null is not that gene doping is undetectable but that it does not work well enough to be worth the risk.

Established Pharmacological cognitive enhancement has one large replicated result and it is negative. Owen et al., in Nature in 2010, ran a six-week online trial with 11,430 completers from 52,617 registrants, mean 24.47 sessions: effect sizes on the trained tasks of 0.73 to 1.63, effect sizes on untrained transfer tasks of 0.01 to 0.22, and the conclusion that there was “no evidence for any generalized improvements in cognitive function following brain training.” Established That is the characteristic shape of nearly every cognitive-enhancement claim here: large, real, and confined to exactly what was practised. Established Structural evidence follows the same pattern — Draganski et al. in 2004 found juggling produced a “transient and selective structural change” in grey matter, and the load-bearing word is transient. Frontier On the drugs themselves this brief prints no effect sizes, and the reason is worth stating: the standard reviews — Battleday and Brem’s 2015 systematic review of modafinil in non-sleep-deprived healthy subjects, the Repantis meta-analyses, and Horvath, Forte and Carter’s 2015 quantitative review titled No Evidence of Cognitive Effects in Healthy Populations From Single-session Transcranial Direct Current Stimulation — could not be retrieved for this brief, and quoting numbers from memory is how this field’s misinformation is manufactured. Frontier What can be said from the record is directional: benefits concentrate in complex executive tasks in rested subjects, are inconsistent on simple attention and memory, and no review in circulation claims a general uplift.

Established Two-thirds of the education polygenic score is not the child’s own genes, and the field’s own consortium reports the number. Okbay et al. 2022, the fourth educational-attainment genome-wide association study, ran on N = 3,037,499 and produced a polygenic index explaining 15.8% of variance in years of education in Add Health and 12.0% in the Health and Retirement Study, a sample-size-weighted mean of 13.3%. Established The same paper reports that the ratio of direct to population effect estimates is 0.556 (s.e. 0.020), and states the consequence explicitly: 0.556² = 30.9% of the index’s R² is due to its direct effect. Established Do the arithmetic on the headline number and the 13.3% becomes about 4.1 percentage points of directly heritable, selectable signal; the remaining nine points are assortative mating, population structure and parental environment correlated with parental genotype. Established Independently, Howe et al. 2022 ran within-sibship association analyses on 178,086 siblings in 77,832 sibships across 19 cohorts and measured how far population estimates shrink when the comparison is between brothers and sisters: educational attainment 47% (41–52%), cognitive ability 22% (6–37%), height 10% (8–12%), smoking 19% (9–30%), body-mass index limited evidence of any difference. Established The correction is directional and it points the wrong way for enhancement: the shrinkage is largest for exactly the traits enhancement advocates care about and smallest for the trait nobody wants to select on.

Established The peer-reviewed gain from embryo selection is small, and its prediction interval spans zero. Turley et al. 2021, in the New England Journal of Medicine, modelled selecting the top-scoring of ten embryos against choosing at random. For European-ancestry couples the expected gain was 0.53 years of education, with a 95% prediction interval of −3.2 to +4.2 years; for height, about 2.5 cm. Established For African-ancestry couples the same procedure returned 0.23 years, roughly 43% of the European figure, with non-European groups performing 25–65% worse across all the conditions studied. Established The disease numbers are worth carrying alongside because they show the same asymmetry: top-of-ten selection gave absolute risk reductions of 0.12 percentage points for type 1 diabetes (35% relative), 5.5 points for type 2 diabetes (16% relative), 1.9 points for breast cancer (15% relative), and 1.8 points for idiopathic short stature — where the relative reduction is a spectacular 77% and the expected height gain is those same 2.5 cm. Frontier Sasha Gusev’s 2025 technical critique puts an upper bound of about 6 IQ points and roughly 10 months of extra schooling on selection from ten embryos, working from within-family heritability ceilings of about 4% for educational attainment and 14% for IQ, and reports that selecting on educational attainment raises bipolar-disorder relative risk by about 1.4× at ten embryos. Frontier Note what the interval means for a family rather than a population: for any individual couple, selection on this score is close to a coin flip.

Frontier The strongest cognitive predictor in existence is a commercial product, and its validation is unreplicated. Herasight states that its CogPGT 1.0 predictor explains 16.4% of fluid-intelligence variance among unrelated UK Biobank individuals and correlates 0.456 with latent general cognitive ability within 4,642 UK Biobank sibling pairs, 0.448 when meta-analysed with the ABCD cohort. (interested — these are the vendor’s own figures.) Established Published academic scores are substantially weaker: Savage et al. 2018, on 269,867 individuals, reported SNP heritability of 0.19 and out-of-sample polygenic prediction of up to 5.2% of variance. Frontier A stronger predictor should give a larger gain than Turley modelled, and it probably does — but no within-family randomised gain estimate has been published for it by anyone, and until one is, the difference between 5.2% and 16.4% is a claim about a score, not a claim about a child. Frontier Orchid Health markets embryo reports claiming to read “99% of an embryo’s genome — 100x more data than standard PGT” across “400+ genetic conditions,” including autism, Alzheimer’s disease, ALS, schizophrenia and coronary artery disease, without publishing sensitivity or specificity for any of them. (interested.) Established Meanwhile the consumer genetics that sits adjacent to this market is straightforwardly dead: Border et al. tested 18 historical candidate genes, 16 polymorphisms and four imputed variable-number tandem repeats for depression across 621,214 individuals and concluded that the earlier positive findings “were false positives,” recommending the field abandon them — which takes COMT, BDNF, 5-HTTLPR and DRD4 panels with it.

Established The enhancements that demonstrably work on healthy adults today are worn, not inherited. Slade, Kochenderfer, Delp and Collins reported in Nature in 2022 that personalised ankle exoskeleton assistance reduced energy per distance by 17 ± 5% in naturalistic outdoor walking and 23 ± 8% on a treadmill at 1.5 m/s, increased self-selected walking speed by 9 ± 4%, and did the personalisation in 32 minutes against 128 for metabolic-cost methods, with ten untethered and nine tethered participants. Frontier Larger figures exist and come with a caveat that must travel with them: Luo et al. 2024 reported metabolic-rate reductions of 24.3% walking, 13.1% running and 15.4% stair climbing from a controller learned entirely in simulation, and Nature attached an editor’s note on 27 July 2026 stating that concerns had been raised regarding the availability of data and code required to support the results and conclusions. Established For the biological ceiling, the best calibration is a real fixed human adaptation: Huerta-Sánchez et al. 2014 showed the Tibetan high-altitude EPAS1 haplotype is Denisovan-introgressed, common in Tibetans and rare in Han Chinese, and associated with lower haemoglobin at altitude. Frontier That is what tens of thousands of years of the strongest available selection pressure buys at one locus: not more of the obvious quantity, but a regulatory fix that avoids a cost. Proposals promising more of the obvious quantity should be read against it.

Frontier Somatic editing has become enhancement-adjacent in practice rather than in principle, because it now moves a continuous physiological variable permanently with one dose. Verve Therapeutics’ VERVE-102 — an adenine base editor and guide RNA in a GalNAc-targeted lipid nanoparticle, aimed at PCSK9 in the liver — reported in the Heart-2 phase 1b, data cutoff 13 March 2025, n = 14 across three cohorts: mean LDL cholesterol reductions of 21% at 0.3 mg/kg (n = 4), 41% at 0.45 mg/kg (n = 6) and 53% at 0.6 mg/kg (n = 4), with mean PCSK9 protein reductions of 46%, 53% and 60%, a maximum individual LDL reduction of 69%, no treatment-related serious adverse events and one grade 2 infusion reaction. (interested — company release.) Frontier The indication is disease, but the template is enhancement: a single dose that resets a lifelong set-point in a tissue, with a dose–response curve rather than an on/off switch.

Frontier Bespoke editing has collapsed the design-to-dose timeline, which changes the economics of any personalised intervention. KJ Muldoon, born 1 August 2024 with CPS1 deficiency, received a custom adenine base editor delivered as messenger RNA in lipid nanoparticles, first dosed 25 February 2025, three infusions over seven weeks, discharged 2 June 2025; design to manufacture took about six months against an estimated eighteen. Frontier That is n = 1 and the indication is a lethal metabolic disease, so nothing about safety generalises — but the manufacturing precedent is the one that matters for enhancement, because enhancement is by definition unpooled demand.

Frontier Restoration keeps overshooting into something that is not restoration. Regeneron’s DB-OTO otoferlin gene therapy, reported in the New England Journal of Medicine in October 2025 via the company’s release, treated 12 children aged 10 months to 16 years, nine unilaterally and three bilaterally: 14 of 15 treated ears improved, nine reached 70 dBHL or better at week 24, and three achieved normal hearing sensitivity, with responses within weeks and stability or continued improvement out to 72 weeks. (interested.) Frontier Lorach et al. 2023 implanted a 38-year-old with C5/C6 tetraplegia with two 64-electrode epidural cortical implants and a 16-electrode lumbosacral paddle; after 40 rehabilitation sessions the participant walked with crutches with the interface switched off, and his WISCI II score rose from 6 to 16 after a three-year plateau on stimulation alone. Frontier The off-state recovery is the interesting part: a device produced a durable biological change it was not designed to produce. Speculative If that generalises, the boundary between assistance and enhancement is not a line but a ratchet.

Frontier Bandwidth into and out of the brain is now measured, and the numbers are still restorative. Willett et al. 2023 achieved 62 words per minute of decoded speech in one participant with ALS using four microelectrode arrays in areas 6v and 44 — 3.4 times the previous record — at a 9.1% word error rate on a 50-word vocabulary and 23.8% on a 125,000-word vocabulary. Established Medtronic’s BrainSense Adaptive deep brain stimulation was approved by the FDA on 24 February 2025, the first closed-loop system to self-adjust stimulation to individual brain activity in real time, with an Electrode Identifier the company says cuts initial programming time by 85%. (interested.) Speculative Reading these as enhancement platforms is a projection, not a finding: every one is indicated for a deficit, and none has been shown to take a healthy person above baseline.

Frontier The live commercial frontier is selection, and its central quantity has never been measured prospectively. Embryo-level polygenic reports are sold today for cognitive ability and for hundreds of conditions; no vendor has published a within-family randomised outcome, and no cohort of selected children has been followed to an outcome by anybody. Speculative The experiment that would settle it is described in section 6 and is not being run, for reasons that are partly ethical and mostly commercial: a null result destroys a product, and there is no regulator requiring the trial. Handwave Meanwhile the far end of the field — pharmacological reopening of developmental critical periods, iterated embryo selection, multiplex germline editing — consists of proposals whose most advanced human demonstration is a 24-participant crossover trial whose second arm failed.

Established The workback chain to a durable, non-trivial enhancement of a healthy human capability has seven links, and the binding two are not scientific. Link one: establish that any target trait has a large-effect, low-pleiotropy human variant. Frontier That is currently false for cognition — within-family signal shrinks by 22% for cognitive ability and 47% for education, and the architecture is thousands of alleles of tiny effect rather than a handful of usable ones. Speculative Link two: show the variant is causal rather than tagging something else, which requires editing it into human tissue and measuring the phenotype, and has not been done for a single cognitive variant.

Frontier Link three is delivery, and it is an industrial constraint with a precise shape: liver is nearly solved and nothing else is. The entire in vivo editing record that reaches therapeutic fractions is hepatic, because GalNAc-conjugated lipid nanoparticles find hepatocytes. Frontier Muscle and central nervous system have no equivalent vehicle; every non-liver programme reaches its target by stereotactic injection, cochlear injection or topical application to an epithelium. Speculative Since the traits enhancement wants are muscular and neural, the tissue where delivery works is the tissue enhancement least wants.

Frontier Link four is durability without hidden damage, and the failure modes are documented in approved products rather than hypothesised. Cas9 double-strand breaks resolve into deletions extending over many kilobases that standard genotyping misses because it examines only the immediate vicinity of the target site. Frontier A five-year follow-up with long-read sequencing and clonal tracking is the minimum readout, and for an intervention in a healthy person the minimum is arguably a decade. Speculative Link five is additivity: no one has demonstrated that dozens or hundreds of small-effect edits sum to their predicted total in any mammal, and accumulated damage plus epigenomic drift are the stated failure modes of the people who propose doing it.

Established Link six is a regulatory pathway for a non-disease indication, and there is none. Every approved editing or gene-therapy product is indicated for a disease; there is no approval route for a product whose endpoint is that a healthy person becomes better than average. Established Link seven is heritability, which requires germline modification: prohibited for reproduction in 70 countries explicitly, permitted with exceptions in five more, and permitted for reproduction in none. Frontier Six and seven are jointly binding, and they are institutional rather than technical — the scientific links have visible experiments attached and the regulatory ones have no experiment attached at all, because they are decisions. Speculative A brief that treats enhancement as a technology-readiness problem has misidentified the constraint. Frontier There is also a mundane practical bottleneck that dominates the arithmetic in section 2: selection gains are quoted from ten embryos, and a typical IVF cycle yields far fewer than ten euploid embryos, so the modelled gain is an upper bound most patients cannot reach.

Frontier This subject consumes results it does not produce, and the first is statistical power in the right design. Within-sibship genome-wide association at roughly ten times current N is what would establish whether the surviving direct effect on cognition is large enough to select or edit on; Howe et al.’s 178,086 siblings is the current state and its cognitive-ability confidence interval, 6–37%, is too wide to plan against. Speculative Nothing about that experiment is conceptually hard; it is cohort assembly and consent, which is why it is a dependency rather than a bottleneck.

Frontier The second is causal validation machinery that belongs to Genetic Engineering. Turning an association into an intervention requires editing candidate variants into human induced-pluripotent-stem-cell-derived tissue and measuring a phenotype, at a throughput that lets thousands of small-effect variants be screened rather than dozens. Speculative For cognition the phenotype itself is the problem: there is no organoid assay for general intelligence, and pretending otherwise is where this field usually goes wrong.

Speculative The third is a claim the germline-engineering advocates make against their own interest, and it deserves to be tested. The Berkeley Genomics Project’s survey of methods for strong human germline engineering argues that the bottleneck is not polygenic scores but epigenomic correction in laboratory-derived gametes — that is, whether in vitro gametogenesis can produce human eggs and sperm with correct imprinting. Established By that survey’s own account, neither paternal nor maternal in vitro gamete production has been achieved in human cells at all. Frontier If the epigenomic claim is right, every argument about predictor accuracy is an argument about the wrong link. Speculative The fourth dependency is measurement: a psychometrics that can detect a shift of a fraction of a standard deviation in a single person, prospectively, without practice effects. Nobody has one, and its absence is why enhancement claims are argued rather than settled.

Frontier The single most valuable experiment in this subject is a prospective cohort of polygenically selected children, and nobody is running it. The design is unglamorous: pre-register the predictor, the selection rule and the outcome; recruit couples already undergoing IVF who elect selection; follow the children to a measured cognitive and educational endpoint with a sibling or non-selected comparison. Speculative It would take two decades and it would settle a market that currently sells on modelled expectations. Frontier Its absence is not an oversight — a null result would be commercially fatal, and no regulator requires the trial because the service is sold as information rather than as a medical intervention.

Frontier Second, the within-family GWAS at scale. Sibship designs at roughly ten times Howe et al.’s 178,086 would narrow the cognitive-ability shrinkage interval from 6–37% to something a selection model can use, and would say whether the 22% point estimate or the 37% upper bound is the right planning number. Established This is the one experiment in this brief whose method is entirely conventional.

Speculative Third, causal validation of a named variant. Base- or prime-edit a candidate cognitive or muscular variant into isogenic human induced-pluripotent-stem-cell-derived tissue and measure a phenotype with a plausible relationship to the trait — for muscle this is straightforward and for cognition it is currently impossible, which is itself a finding worth stating. Frontier Fourth, dose-ranging outside the liver: a VERVE-102-style study in skeletal muscle or central nervous system, reporting edited fraction by tissue, is the measurement that would tell us whether somatic enhancement is a real category or a liver-only one.

Speculative Fifth, an additivity test. Multiplex editing of dozens of independent small-effect loci in a long-lived somatic tissue or a primate model, reporting whether the phenotypic change matches the summed predicted effects and what accumulated damage costs. Frontier Sixth, a five- to ten-year safety readout with long-read sequencing for kilobase-scale deletions and clonal tracking for insertional expansions — the insertional-oncogenesis events seen in approved gene therapies appeared over a span of years rather than months, which is longer than most trials watch. Frontier Seventh, and cheapest: a randomised trial of a candidate cognitive enhancer in healthy adults powered for a transfer endpoint rather than a trained-task endpoint, pre-registered, with a null hypothesis the sponsor can lose. Owen et al. showed in 2010 what happens when the transfer endpoint is measured, and almost nothing in the nootropics literature since has measured it that way.

Frontier The engineering that has actually shipped in this subject is mechanical, and its distinguishing feature is per-person optimisation in minutes. The exoskeleton result that matters is not the 17% metabolic saving but the 32-minute personalisation: a controller fitted to one individual’s gait in the time it takes to walk around a block, rather than in a two-hour laboratory session. Frontier That is a different scaling law from anything biological — it improves with data and compute rather than with biology, and it is reversible by taking the device off, which is why it clears a risk bar no genetic intervention can currently clear.

Frontier On the biological side the engineering requirement is a delivery vehicle, not an editor. Editing chemistry is the least binding link: adenine base editors delivered in GalNAc lipid nanoparticles already produce dose-dependent, apparently permanent changes in a human physiological variable. Speculative What does not exist is the equivalent vehicle for skeletal muscle or brain at therapeutic edited fraction, and no published roadmap in this brief’s sources gets there by a stated date. Frontier Manufacturing is the second requirement and it has moved fast: a patient-specific editor went from sequence to first dose in about six months against an eighteen-month estimate, which is the precedent that makes bespoke, single-person interventions conceivable at all.

Frontier The third engineering requirement belongs to the adversary side: detection. Gene-doping surveillance runs on PCR against exon–exon junctions in complementary DNA and on next-generation sequencing, and both are defeated by the same thing — a construct that looks like natural variation, or a natural variant that looks like a construct. Speculative A detection method that keeps pace with editing would need to be longitudinal rather than cross-sectional: a stored personal genomic and transcriptomic baseline against which later samples are compared, which is a data-governance problem as much as an assay problem. Handwave No sporting body has proposed it at scale, and the athlete-privacy objection to holding baseline genomes for a career is not obviously answerable.

Established This subject sits downstream of the technology and upstream of the argument. Genetic Engineering supplies every tool named here and owns the delivery, on-target-damage and manufacturing constraints; nothing in enhancement moves until those do. Frontier Neurogenetics owns the quantitative core of the cognitive claim — the polygenic architecture, the within-family corrections, and the gap between twin heritability and SNP heritability that governs how much there is to select on.

Frontier Neuroplasticity Engineering holds the only mechanism in this brief that would count as genuine headroom rather than reallocation: reopening a developmental window in an adult brain, with the valproate absolute-pitch trial as its single direct human test. Frontier Cognitive Enhancement is the same subject approached from the intervention side rather than the substrate side, and the two briefs should be read as one argument. Speculative Brain-Computer Interfaces supplies the bandwidth numbers that any claim about augmented cognition has to beat, and currently they are restorative rather than superhuman.

Frontier Genetic Engineering also owns the negative result that most constrains this brief’s ambitions: the on-target damage literature, where Cas9 breaks resolve into kilobase-scale deletions that routine genotyping never sees. Frontier Precision Medicine shares this subject’s central methodological problem — polygenic scores sold ahead of their validation — and its ancestry-transportability failure is the same failure measured on a different endpoint. Speculative Human Adaptation for Space is where enhancement acquires its only serious mission rationale, since a crew that cannot be resupplied is the one population for which intervening in a healthy person has a stated justification. Speculative Artificial Wombs and Longevity Therapies bracket the subject at either end of a life: the first would remove the embryo-count constraint that caps selection gains, and the second is enhancement of a healthy person by another name, with the same missing risk bar.

Established Heritable enhancement is not unregulated; it is prohibited nearly everywhere, and the prohibition is unusually uniform. The published survey of national law records 70 countries with explicit bans on germline modification for reproduction, five more permitting it only with exceptions, and no country permitting it for reproduction. Established Somatic enhancement occupies the opposite position: there is no prohibition and no pathway, because every approval framework in existence attaches a product to a disease indication. Frontier The regulatory position on somatic enhancement is not permissive; it is silent by exclusion, and that silence is what the embryo-selection market operates in, since selecting among embryos modifies nothing and is regulated, where it is regulated at all, as an information service.

Established The funding is concentrated, named, and largely philanthropic rather than commercial. Reporting by the Center for Genetics and Society names in vitro gametogenesis startups Gameto, Ivy Natal and Conception; records that Good Ventures, the Cari Tuna and Dustin Moskovitz foundation, gave $6.5 million to a team that carried out in vitro gametogenesis in mice; that Open Philanthropy funded reproductive-biotechnology research and the institute Nick Bostrom directed; and that Jaan Tallinn invested in Conception, with George Church advising Gameto. (interest against — the author is a critic of the field, and the facts are reported as such.) Frontier A field whose capital comes from a small number of ideologically aligned donors has a specific failure mode: the research agenda tracks the donors’ prior beliefs, and disconfirming results have no natural sponsor.

Established Sport is the only institution that has actually written enhancement rules, and its record is instructive. WADA prohibited gene doping in 2003 on a precautionary basis, before any capability existed, and in twenty-three years has detected nothing. Frontier That is either a triumph of deterrence, a failure of detection, or evidence that the technology never worked — and the three are not currently distinguishable, which is itself a governance finding. Speculative The institution that does not exist is the one that would set a risk threshold for intervening in a person who is well. Frontier Every regulator in medicine sizes acceptable risk against the harm of an untreated disease; remove the disease and the denominator goes to zero, and no agency has published a framework for what to do then.

Speculative The hardest ethical problem here is not fairness or hubris; it is that the standard risk calculus divides by zero. Medical ethics prices an intervention’s risk against the burden of the condition it treats. For a healthy subject there is no burden, so the acceptable-risk threshold has to be set by something else — a floor of absolute safety that no genetic intervention can meet, a competence-based consent standard borrowed from cosmetic surgery, or a societal judgement about which capabilities are worth risk. Frontier None of the three has been written down by a regulator, and the absence is the reason somatic enhancement has no approval pathway rather than a restrictive one.

Speculative The strongest principled objection is worth stating at full strength even though its primary text could not be obtained for this brief. Michael Sandel’s argument from “giftedness” holds that the drive to mastery over our own natures corrodes humility, solidarity and responsibility, and that the objection stands even if enhancement is safe, fair and effective — that is, it is not a risk argument dressed up. Frontier The reply the mainstream makes is that the same argument condemns education, orthodontics and vaccination, and that no principled line separates them; the reply to the reply is that the line is not principled but historical, and that a practice’s having been normalised is not an argument. Established The President’s Council’s own report carries a version of the giftedness argument, and also its own concession that the categories it needs do not hold.

Established The measured distributive fact is sharper than the speculative ones. The same embryo-selection product returns 0.53 years of education for European-ancestry users and 0.23 years for African-ancestry users — about 43% of the benefit — because the training data are European. Frontier That is not a projected inequity; it is a measured one, present at the point of sale, in a product being marketed now. Speculative The second distributive problem is positional: for traits that matter mainly by rank — height, examination performance, athletic placing — a universal enhancement changes the distribution and not the ranking, so the benefit to the buyer depends on other people not buying. Handwave Whether a society ends up with universal access, an arms race, or a prohibition that only the mobile can evade is not a technical question and no one has a model of it worth quoting.

Speculative If enhancement works, its first-order civilizational effect is a change in the variance of human capability, not the mean. Every measured intervention here is small, expensive and individually purchased, which shifts the top of a distribution rather than lifting it — and the traits people want most are the ones where the within-family signal is weakest, so the shift is smaller than the price suggests. Frontier A society that pays a great deal for fractions of a standard deviation has made a distributive choice, not a developmental one.

Speculative The second-order effect runs through legitimacy rather than biology. A market selling cognitive selection on scores whose direct component is roughly a third of the headline figure is generating a widely believed factual error about what genes do, and that error has political uses independent of whether the product works. Frontier The measured ancestry gap gives that error a direction: any enhancement built on current polygenic scores delivers systematically less to non-European populations, which converts a technical limitation into a distributive one at population scale.

Handwave The long-horizon question is what a genuinely enhanced human would be measured against. Every instrument in use — IQ, years of education, VO2 max, bone density — is normed on the unenhanced population, so a large enough shift destroys its own yardstick, and the comparison that would matter is against a counterfactual version of the same person that cannot be observed. Handwave If enhancement ever becomes general, the honest description of its effect will not be a number but a change in what the species treats as normal — which is precisely the concept the President’s Council found it could not define.

These horizons track two clocks that are not coupled: a scientific one, where the experiments are specified and mostly unfunded, and an institutional one, where the binding constraints are decisions nobody has been asked to make.

• 10 yr: Frontier Embryo polygenic selection continues to be sold and remains unvalidated by any prospective outcome study; the market grows on modelled expectations. Frontier Within-sibship GWAS gets larger and the cognitive-ability shrinkage interval narrows, most likely confirming that the direct effect is real, small, and highly polygenic. Frontier Somatic base editing acquires a second solved tissue after liver, or conspicuously fails to. Speculative Personalised exoskeletons and closed-loop neuromodulation continue to be the only interventions that measurably improve a person’s function this decade, and both are devices. Speculative Gene doping is still not detected, and the argument about why remains unresolved.
• 25 yr: Speculative A regulator somewhere is forced to rule on a non-disease indication — most plausibly for a longevity endpoint or a space crew, not for cognition — and that ruling, whichever way it goes, becomes the framework everything else is argued against. Speculative In vitro gametogenesis in human cells either works or is shown to fail on imprinting, which decides whether iterated embryo selection is a real proposal or a permanent thought experiment. Speculative Multiplex editing is demonstrated at tens of loci in a mammal, and the additivity question gets its first real answer. Handwave If the answer is that effects do not add, most of the germline-enhancement literature becomes obsolete in a single result.
• 50 yr: Speculative Heritable enhancement, if it happens at all, happens in one jurisdiction that decides to permit it, and the resulting cohort is small, watched and contested. Handwave Reopening developmental plasticity in adults is the only route to cognitive headroom that does not run through selection arithmetic, and on current evidence it rests on a single 24-participant trial whose crossover arm failed. Handwave Measurement, not biology, becomes the limiting discipline: distinguishing a real half-standard-deviation gain from practice effects and selection into treatment is harder than producing one.
• 100 / 250+ yr: Handwave On this horizon the question stops being whether a capability can be raised and becomes whether the concept of a baseline survives, since every instrument is normed on the population being changed. Handwave The honest terminal position is the one the evidence supports today: no intervention in the fetched record has moved a healthy adult beyond the existing human range on a durable basis, and a century of that record would be a finding about human biology rather than about effort.

• Depends on Depends on Genetic Engineering for every tool this subject would use and for all three of its unsolved technical links: a delivery vehicle that reaches muscle or brain at therapeutic edited fraction, an editing chemistry whose on-target damage is characterised at the kilobase scale rather than at the target site, and a manufacturing route that makes a one-person intervention affordable. None of those is an enhancement problem; they are all inherited from the therapeutic programme, and enhancement cannot move ahead of them because it has a strictly higher safety requirement and a strictly smaller patient population to amortise across.
• Requires (not on this map) Two constraints govern this subject and neither is a laboratory result. The first is a published threshold for how much risk may be imposed on a person who is not ill. Medicine prices risk against the burden of a disease, and with the disease removed the denominator vanishes; no agency anywhere has written the replacement, which is why somatic enhancement has no approval pathway rather than a restrictive one, and why the market has organised itself around embryo selection, which modifies nothing and is regulated as information. The second is a detection capability that tracks editing rather than trailing it. Gene doping was prohibited in 2003 and nothing has ever been detected, and the assays in use are defeated equally by a construct that mimics natural variation and by a natural variant that mimics a construct — so the only method that would work is longitudinal comparison against a stored personal baseline, which is a data-governance decision that no sporting or medical body has taken.
• Enables Enables very little that is downstream in the ordinary sense, and this is a real feature of the subject rather than an omission: enhancement is a terminal application, not an input to another technology. What it does supply is a forcing function. A regulator obliged to set a risk threshold for a healthy subject would produce a framework that Longevity Therapies needs for exactly the same reason and does not have, and that Human Adaptation for Space would need before any crew modification could be proposed. The measurement instruments this subject lacks — assays sensitive to fractional-standard-deviation change in one person — would be immediately useful to Precision Medicine.
• Adjacent Adjacent to Neurogenetics, which owns the polygenic arithmetic this brief spends most of its length on; to Cognitive Enhancement, the same argument from the intervention side; to Neuroplasticity Engineering, which holds the one candidate mechanism for genuine headroom rather than reallocation; to Brain-Computer Interfaces, whose measured bandwidth is the benchmark any augmented-cognition claim must beat; and to Artificial Wombs, which would relax the embryo-count limit that caps every selection estimate in section 2.

Frontier “Embryo selection buys 8.5 IQ points.” The number is a range across three embryos, not a gain from choosing one. Herasight’s own marketing states that its CogPGT 1.0 predictor is “able to detect on average an 8.5 IQ point difference between just three embryos.” (interested.) That is the expected maximum-minus-minimum spread among three siblings on the predictor scale — the width of the distribution you are choosing from — and not the expected gain from choosing the top one over choosing at random, and not a measured phenotypic outcome in a child. Established The peer-reviewed estimates of gain are these: Turley et al. 2021 give 0.53 years of education from selecting the top of ten embryos in European-ancestry couples, with a 95% prediction interval of −3.2 to +4.2 years, and about 2.5 cm for height; the same procedure returns 0.23 years for African-ancestry couples. Frontier Gusev’s upper bound, worked from within-family heritability, is about 6 IQ points from ten embryos. Established Line those up: a spread quoted for three embryos exceeds the most generous published upper bound for ten, and the two figures are not the same quantity. Frontier This is the single most distorted number in the subject area, and it propagates because it is technically true of the predictor and says nothing about the child. Frontier One publication would change the picture — a within-family randomised gain estimate for the commercial predictor — and the vendor has not produced it.

Established “Polygenic scores explain 13–16% of variance in education.” True, and two-thirds of it is not the child’s own genes. Okbay et al. 2022 report a polygenic index explaining 15.8% of variance in Add Health and 12.0% in the Health and Retirement Study, weighted mean 13.3%, on N = 3,037,499. Established The same paper reports a direct-to-population effect ratio of 0.556 (s.e. 0.020) and states the implication in its own words: 0.556² = 30.9% of the index’s R² is due to its direct effect. (interest against — this is the consortium that built the score reporting that roughly 69% of its predictive power is not direct genetic effect.) Established So the selectable component of that 13.3% is about 4.1 percentage points; the rest is assortative mating, population stratification, and parental environments correlated with parental genotype — real prediction, useless for intervention. Established Howe et al. 2022, on 178,086 siblings in 77,832 sibships, get there independently: within-sibship estimates shrink population estimates by 47% (41–52%) for educational attainment, 22% (6–37%) for cognitive ability and 10% (8–12%) for height. Frontier Quoting 13% for education without the 0.556 ratio overstates the heritable, selectable component by roughly a factor of three, and the shrinkage is largest for exactly the traits enhancement wants and smallest for the one nobody proposes to select on.

Established “The therapy/enhancement line is clear.” The President’s Council on Bioethics drew it, built a report around it, and concluded it is “finally inadequate to the moral analysis.” Frontier Growth hormone for short stature is the report’s own demonstration: identical intervention, identical outcome, sorted by the availability of a diagnosis. Speculative Every proposal to regulate enhancement by defining it inherits that failure, which is why the operative line in practice falls between modifying an embryo and choosing among embryos — technique rather than purpose.

Frontier (Enthusiast error) “Nootropics have solid meta-analytic support.” The strongest replicated finding in healthy cognitive enhancement is an absence: Owen et al. 2010, on 11,430 completers, measured trained-task effect sizes of 0.73–1.63 and transfer effect sizes of 0.01–0.22, and reported “no evidence for any generalized improvements in cognitive function.” Frontier The pharmacological reviews people cite here — Battleday and Brem on modafinil, Repantis and colleagues on stimulants, Horvath and colleagues on single-session transcranial direct-current stimulation — could not be retrieved for this brief, and no effect size from any of them is printed anywhere in it. Established What is safe to say is that the tDCS review’s own title reports no evidence of cognitive effects in healthy populations from a single session, and that no review in circulation claims a general uplift. Speculative A drug producing general transfer would be the most surprising result in cognitive science — a reason to demand a pre-registered transfer endpoint, not to disbelieve in advance.

Established (Enthusiast error) “A genetic panel can tell you your COMT and BDNF profile and what to take for it.” Border et al. 2019 tested 18 historical candidate genes, 16 polymorphisms and four imputed variable-number tandem repeats across 621,214 individuals and found essentially nothing — one marginal COMT effect, a DRD2 signal that failed to replicate, no gene-by-environment interactions — concluding that the prior positive findings “were false positives.” Frontier The candidate-gene era was not underpowered and broadly right; it was wrong, and products still sold on those genes are selling a literature that has been formally retired.

Established “Gene doping is already widespread; they just cannot catch it.” Prohibited since 2003; zero confirmed cases in competition in twenty-three years; the only court evidence anywhere is an email in which a coach attempted to buy an EPO construct. Speculative The undetectability reading requires believing that a technology whose therapeutic version struggles with delivery and durability in hospitals works reliably in clandestine use. Frontier The alternative — that it does not work well enough to be worth the risk — fits the same evidence and makes fewer claims.

Established “The myostatin boy proves muscle enhancement is a solved single-gene problem.” He was homozygous for a splice mutation producing no detectable mature myostatin, with quadriceps 7.2 standard deviations above the control mean; his heterozygous mother was merely athletic. Frontier The phenotype needed complete pathway loss, the heterozygous state buys little, and one child is not a safety dataset. Frontier The drug programmes on the same target have added lean mass without proportional function.

Established “Exoskeletons do not really help.” They do, and the numbers are specific and outdoors: 17 ± 5% less energy per distance, 9 ± 4% faster self-selected walking, personalised in 32 minutes. Frontier The larger simulation-learned figures — 24.3%, 13.1%, 15.4% — carry a Nature editor’s note of 27 July 2026 on the availability of data and code, and should not be quoted without it.

Established “Enhancement is unregulated.” Heritable enhancement is prohibited for reproduction in every country surveyed — 70 with explicit bans, five permitting only with exceptions, none permitting it — and somatic enhancement has no approval pathway because every approved product carries a disease indication. Frontier The position is not permissive; it is silent by exclusion, and silence is what the embryo-selection market is built in.

Speculative “Evolution already optimised us, so enhancement is futile.” Bostrom and Sandberg’s Evolutionary Optimality Challenge asks of any proposal: if this would be an improvement, why have we not already evolved it? Speculative They name three legitimate answers, and the three double as a filter. Changed tradeoffs — caloric abundance makes a costlier brain affordable; modern literacy demands exceed anything ancestral. Value discordance — we optimise for things other than inclusive fitness, with contraception as the completed case. Evolutionary restrictions — local optima and missing tools, such as antagonistic pleiotropy at APOE, or supplying lactase pharmacologically to populations without the recent allele. Frontier A proposal fitting none of the three should be disbelieved; one that fits has answered the strongest general objection and still owes the specific ones.

Speculative “Iterated embryo selection is nearly here.” The idea has been independently reinvented at least five times — velogenetics in 1991, whizzogenetics in 1998, iterated embryo selection in 2009, in vitro eugenics in 2013, in vitro breeding at least three times in 2018 — and the livestock literature claims it could be 30 to 40 times more efficient than conventional breeding. Established What has been demonstrated in humans is IVF, preimplantation diagnosis and stem-cell derivation; what has never been demonstrated in any mammal is a complete iterative cycle producing viable offspring across multiple rounds. Frontier The most-cited academic treatment of germline cognitive selection could not be read for this brief and no figure from it appears here — worth saying plainly, because its numbers are widely repeated at second hand.

Handwave The terminal position, stated as plainly as the evidence allows. Every enhancement in this brief that has actually been measured is one of three things: a restoration of a lost function, an externally worn device, or a statistical shift of a fraction of a standard deviation within a family. Speculative No intervention in the retrieved literature has moved a healthy adult human capability beyond the existing human range on a durable basis. Frontier That is a strong claim and it survives the evidence assembled here. Handwave It is not a prediction that it cannot be done — the Evolutionary Optimality Challenge names three routes by which it might be — but anyone asserting that it has been done, or is about to be, owes a number with a prediction interval attached, and that is exactly what this subject’s most-quoted figures do not have.