Ecosystem Gene Drives

Speculative If a homing drive behaves in an open population the way it behaved in a 4.7 cubic-metre cage — which no release anywhere on Earth has tested, and which section 5 says nothing currently measures — the prize is a malaria vector that stops existing rather than one that is held down. That distinction is the whole case. Every other vector intervention buys a season of reduced biting for a season of spending; a drive is the only proposed intervention whose unit of action is an allele frequency in a wild population rather than a quantity of product delivered to a household, so its cost does not scale with area covered or years run.

Established The baseline it would be measured against is 610,000 malaria deaths in 2024, about 95 per cent of them in Africa, with children under five making up roughly three-quarters of the African toll. The existing toolkit is not failing in aggregate — nets, spraying, diagnostics and drugs are credited with averting 2.3 billion cases and 14 million deaths since 2000 — but it is losing ground in places: 2.9 million more cases in Ethiopia and 1.9 million more in Madagascar between 2023 and 2024, under insecticide resistance and an invading urban vector. A drive is proposed as the tool that does not have to be re-bought.

Established In cages the effect size is total rather than marginal, and that is what makes the proposition worth arguing about. Two cages seeded with 600 insects each went from a drive in a quarter of the males to no eggs at all by generation 8 and generation 12; six larger cages of about 570 adults were emptied in 245 to 311 days from a starting allele frequency of 12.5 per cent. Nothing in the conventional toolkit produces a number of that shape, and nothing in it is irreversible either — the two facts are the same fact.

Frontier The upside is also unusually badly distributed, and that belongs here rather than only in section 11. The gain would accrue to rural households in a handful of African countries; the risk of a mistake accrues to everyone sharing the mosquito population, which is a different and larger map; and the decision sits with national biosafety authorities that, on the published survey evidence, mostly have no gene-drive-specific provisions at all. The condition on all of it is one missing number: the effect of a drive on malaria transmission in an open population, which section 7 says how to get and which nobody has yet been permitted to measure.

Established A gene drive is a genetic element engineered to be inherited by more than half of an organism’s offspring, so that it increases in frequency generation on generation even when it makes its carrier worse off. The engineering object is therefore not an organism but an allele frequency in a wild population, and that is what separates the subject from every other release of a modified organism: a sterile-insect release decays after the last shipment, while a drive is designed so that the last shipment is the only shipment. In laboratory strains, CRISPR-based homing drives are reported at inheritance rates of roughly 95 to 99 per cent against the Mendelian 50.

Established Two purposes, and they carry different risk profiles. A suppression drive spreads a construct that damages fertility or skews the sex ratio, so the population shrinks and, in the cage results, disappears. A modification drive leaves the population in place and spreads a trait making it a poor host for the parasite. Suppression removes a species from a local food web; modification does not, and its critics argue it leaves a permanently altered wild genome behind instead. This brief treats both, because the governance questions are the same and the ecological questions are not.

Established Drives divide again on whether they need a threshold. A non-threshold (self-sustaining) drive is expected to spread from an arbitrarily small introduction, which makes a single release sufficient and containment a design problem. A threshold drive is one whose spread is predicted to reverse below some starting frequency, so a release under it dies out. Self-limiting architectures — split drives, whose components for spread are deliberately separated, and daisy-chain designs, whose capacity to keep driving is exhausted after a bounded number of generations — are meant to make a release finite in space or time. This brief describes them at that level and no further.

Established Resistance here has a specific technical meaning a policy reader needs. When the drive fails to copy itself into the matching chromosome, the repair can leave a sequence the drive no longer recognises but that still works. Such an allele is immune to the drive and, in a suppression design, is favoured by selection, so it can spread faster than the drive and stop it. The published countermeasure is to aim at a sequence that cannot tolerate change, because a broken version of it is not a survivor. Whether that holds in wild populations carrying far more standing variation than a laboratory colony is one of this brief’s open questions.

Established This brief contains no construct design and no guidance on building or improving a drive, and that is a rule rather than an oversight. The assessment is of ecology, evidence and governance. The vocabulary above is the minimum a regulator or a parliamentary committee needs to read a risk assessment; everything past it is laboratory guidance and out of scope. The general capability this subject consumes is in Genetic Engineering, the wider record of organisms designed to specification in Designer Organisms.

Established The anchor result is real and its usual one-line summary is slightly wrong. Kyrou and colleagues in Andrea Crisanti’s laboratory ran a homing drive against the female-specific isoform of doublesex in Anopheles gambiae in two caged populations, each seeded with 300 wild-type females, 150 wild-type males and 150 drive-carrying males. In cage 2 the drive reached 100 per cent frequency by generation 7 and the cage produced no eggs in generation 8. In cage 1 the drive sat at 65 to 70 per cent through generations 4 to 8, reached 100 per cent at generation 11, and the cage failed in generation 12. The commonly quoted “seven to eleven generations” is the interval to fixation of the drive, not to collapse of the population. This brief was commissioned with the shorter version and reports the correction as a finding.

Established The resistance claim needs one word changed: resistant sequences did arise, but none of them worked. Amplicon sequencing of pooled samples of at least 359 mosquitoes from generations 2 to 5 found low-frequency insertions and deletions at the target site, up to 1.16 per cent among non-drive alleles, none encoding a functional transcript and none showing any sign of positive selection. The correct statement is not “no resistance emerged” but “resistance emerged and was not viable” — a stronger result and a more fragile one, because it rests on the target being intolerant of change in wild genetic backgrounds as well as in a colony.

Established The result replicates at larger scale under conditions built to be less forgiving. Hammond and colleagues ran the same class of drive in six cages of 4.7 cubic metres, each holding about 570 adults in age-structured, overlapping-generation populations with artificial dawn and dusk, swarming arenas and oviposition sites. Every cage was eliminated: 245 to 311 days from a starting drive frequency of 12.5 per cent, 266 to 276 days from 25 per cent. The authors report that no mutant alleles were generated that could conceivably code for a functional DSX protein, and say in the same paper that comprehensive resistance testing and environmental risk assessment will be needed ahead of field trials.

Established A second architecture reaches the same endpoint faster. Simoni and colleagues combined that target with a sex-distorting element, obtaining a 93.1 per cent male bias in the offspring of heterozygous males; caged populations collapsed at generations 9 to 13 from a 2.5 per cent start and at generations 5 to 6 from 25 per cent, with no functional resistance at the target. This matters for policy because it shortens the window between release and effect, and therefore the window in which anything could be reconsidered.

Frontier Outside that one exceptionally conserved target, resistance is a live problem and the published record says so. A population-modification drive in Anopheles coluzzii reported by Green and colleagues blocked transmission strongly — 29.7 per cent of mice (11 of 37) exposed to infected homozygous mosquitoes developed parasitaemia against 97.1 per cent (33 of 34) exposed to infected controls — but drive-resistant mutations at the second target persisted in roughly 30 per cent of individuals, a functional mutation arose that preserved viability while blocking homing, and the guide array contracted from six copies to two or five. A drive that decays into a non-driving transgene is not a failed release; it is a permanent alteration of a wild population that no longer does the thing it was released to do.

Frontier A suppression drive in the invasive urban vector Anopheles stephensi is at an earlier stage. Xu and colleagues report roughly 74 per cent inheritance alone, rising to complete conversion when paired with a helper line, and resistance-allele formation of 0.657 and 1.38 per cent. Their projections run 317 weeks and assume repeated releases; the authors make no claim of field readiness.

Established The single most important number in this brief is zero: the number of gene drives ever released into the open anywhere on Earth. Fifteen years after the first proof-of-concept work, no field trial of a gene drive organism has taken place, and the most advanced programmes remain in contained laboratory phases. Every efficacy figure above comes from a cage; every ecological figure below comes from a cage, a mesocosm or a model. That is not a gap to speculate across. It is the finding, and it governs every flag on this page.

Established What has been released is a modified mosquito that does not drive, and it under-performed. In July 2019 Target Malaria released about 6,428 transgenic sterile-male Anopheles coluzzii alongside 8,422 wild-type siblings at Bana in western Burkina Faso, the first release of a genetically modified African malaria mosquito. The modified males had lower survival and were less mobile than their siblings. That is a fitness finding about a strain rather than a drive, and it is the only open-field data the field has on how an engineered Anopheles behaves outdoors.

Established The programme then stopped, and it did not stop for a safety failure. On 6 August 2025 Burkina Faso’s authorities authorised a second release; on 11 August a male-bias, non-driving strain was released at Souroukoudingan; on 18 August facilities were sealed; on 22 August the Ministry of Higher Education, Research and Innovation suspended all Target Malaria activities and ordered remaining samples destroyed. Nature Africa records that the government cited only “national interests” and that there was no official indication of biosafety failures or regulatory breaches; an independent governance analysis lists the reasons given as lack of transparency in protocols, divided expert assessment and ecological doubt about eradicating mosquito populations.

Frontier How many mosquitoes were in that August 2025 release is genuinely disputed, and the brief prints the disagreement rather than an average. One critical account gives 16,000; another gives 75,000; the programme’s own figure was not obtainable here. A field in which the size of the last release before a national shutdown is known only through two opposed advocacy sources differing by a factor of nearly five has a disclosure problem independent of its science.

Established The shutdown propagated. Uganda’s programme had been relying on Burkina Faso for the field data on transport, survival, lifespan and mating competitiveness its own risk assessment would need; its lead investigator is quoted saying that without that data “we now have to generate it ourselves, which takes more time and resources”. The insectary at Bobo-Dioulasso, funded by the Gates Foundation since 2012, closed.

Established International law has moved toward guidance rather than prohibition. At the 2018 Conference of the Parties to the Convention on Biological Diversity, Bolivia, El Salvador, Grenada and Egypt supported a moratorium on gene drive release, backed by La Via Campesina, the Global Youth Biodiversity Network, thirty-six African civil society organisations and Friends of the Earth International; a majority of African countries advocated for gene drives; Brazil and Canada pressed for flexible restrictions; the decision that passed on 29 November 2018 called for precaution and case-by-case assessment. The 2022 decision added free, prior and informed consent of indigenous peoples and local communities. At the 2024 meeting in Cali the parties to the Cartagena Protocol welcomed additional voluntary guidance, established an expert group and mandated capacity-building, with no moratorium language.

Established The conservation sector held its own vote in 2025 and the moratorium lost. At the IUCN World Conservation Congress in Abu Dhabi in October 2025, members adopted Motion 87, the IUCN Policy on Synthetic Biology, and narrowly rejected Motion 133, a moratorium on releasing genetically modified organisms into wild populations. The 88 per cent figure for Motion 87 comes from Revive & Restore (an organisation that campaigns for conservation biotechnology, and therefore an interested party); the outcome is corroborated by critical reporting, the margin is not. Opposition came from the Coordinator of Indigenous Organizations of the Amazon River Basin, Wise Ancestors and Friends of the Earth, on grounds of consent, data sovereignty and irreversibility.

Established Regulatory capacity in the countries that would host a release is documented and thin. A 2024 review of African biosafety frameworks finds Kenya (2022) and Nigeria (2019) with genome-editing-specific instruments, Eswatini, Benin and Uganda applying general biosafety law, and most countries with nothing tailored to genome editing or gene drives; it names limited resources and expertise, poor coordination between authorities, and public resistance reported in nine countries. It also records Uganda and Burkina Faso planning field trials within five to ten years — a plan written before August 2025. Section 10 takes the instruments above them apart.

Frontier On the ecology of removing the vector, the two best available syntheses disagree, and they disagree about evidence rather than values. A review assembled for regulators argues that predation on Anopheles is diffuse and substitutable: mosquitofish fed engineered mosquitoes for seven weeks showed no significant difference in length, mass, survival or sex ratio; damselfly diet studies put mosquito larvae at about 3.5 per cent of total arthropod intake; a swarm of roughly 1,000 mosquitoes is about 0.2 grams of biomass. Against that, a systematic review of 117 gene-drive modelling publications finds most models deterministic, non-spatial and not tailored to a target organism, with mating-related factors and interspecies interactions largely excluded and food-web and ecosystem-service impacts absent altogether. Both can be true: the direct trophic effect may be small and the apparatus may still be incapable of showing it.

Frontier The confinement literature is thinner than its role in the debate suggests. The same review finds only twelve of those publications examining mitigation measures at all — split drives, threshold-dependent systems, self-limiting designs — and concludes that at their current stage of development these have not been sufficiently tested to allow firm conclusions about effectiveness. It adds that the stepwise confinement approach used for conventional modified organisms may not be feasible for gene drive organisms. Confinement is the load-bearing promise in almost every proposed pathway to a first release, and it is supported by twelve modelling papers and no field data.

Established The expert body convened under the Cartagena Protocol reaches the same conclusion in its own language. Its 2024 analysis lists incomplete understanding of off-target effects, long-term ecological impacts and the potential for a drive to evolve resistance; notes that most models address the spread of drives rather than their effect on the environment; and poses the jurisdictional question no instrument answers, whether a risk assessment should be framed by the protection goals of the host jurisdiction or by those of the entire area of potential spread. The guidance adopted alongside it records that some such organisms present a potential inability to halt the spread or reverse its action and effects.

Frontier The disease burden the whole argument rests on is large, and its exact shape depends on which report year you read. The figures used here — 610,000 deaths and 282 million cases in 2024, 95 per cent of cases in Africa, roughly three-quarters of African deaths in children under five, 2.3 billion cases and 14 million deaths averted since 2000 — come from a commentary by Fredros Okumu of the University of Glasgow citing the World Health Organization, with the death figure independently repeated by a second source. The case figure is higher than the series published for earlier years, and this brief reports it as its source states it rather than reconciling two documents it could not both open.

Established The most useful thing this brief can do is separate the disputes about facts from the disputes about values, because they are routinely conducted as one argument and have different remedies. A factual dispute closes on a measurement, and the brief names the measurement for each. A values dispute closes on no measurement, and pretending otherwise is how a technical body ends up adjudicating a question of authority. The Institute takes no position on either kind; it reports which is which.

Frontier Factual dispute one: would a drive that works in a cage work in a wild population? For yes: the target has resisted functional mutation across two drive architectures and two laboratories, in populations with age structure and overlapping generations, and the resistant indels that arose showed no positive selection. For no: cage populations are colonies with depleted standing variation, no immigration, no seasonal collapse and refill, no spatial structure and no competing species; and the only engineered Anopheles ever released outdoors had lower survival and lower mobility than its wild siblings. What discriminates: a release into a bounded wild population with allele-frequency tracking, which is the experiment in section 7. Nothing short of it does.

Frontier Factual dispute two: can a “local” drive be confined? For yes: threshold and self-limiting architectures are mathematically well described, a drive below its threshold is predicted to disappear, and peripheral or island populations can have low enough gene flow to act as boundaries. For no: only twelve of 117 modelling publications examine these measures and the review judges them insufficiently tested; the models are mostly non-spatial, so they cannot answer what they are being asked; and the island premise has been challenged directly in the conservation literature on the ground that gene flow is an empirical property of a population rather than of a coastline. What discriminates: a measured gene-flow map of the candidate population, plus a released self-limiting drive with a pre-registered spread boundary and independent sampling outside it. This is answerable and nobody has answered it.

Frontier Factual dispute three: what happens to the food web when a vector species is removed? For little: predation studies show substitutable prey and negligible biomass, and no obligate predator or pollinator dependency is documented for these species. For unknown: every one of those studies is a laboratory or enclosure study, the review that assembles them says so, and the modelling that would extrapolate them omits food webs entirely. Note the shape — the sceptical position is not that harm is likely but that the apparatus for detecting it does not exist. What discriminates: pre-specified ecological endpoints across a suppression gradient, with baseline data collected for several seasons first. That study has never been funded because there has never been a release to attach it to.

Frontier Factual dispute four: does a moratorium make the eventual release safer or more dangerous? For safer: an irreversible act should not precede the evidence, the instruments admit they cannot presently assess the risk, and a pause costs nothing unrecoverable. For more dangerous: regulators have argued in published commentary that without field trial data risk assessment cannot improve, and without a clearer assessment no authority will approve anything — a loop in which precaution prevents the evidence precaution demands. What discriminates: partly a measurement, whether staged confined releases reduce assessment uncertainty; partly nothing, because who carries the burden of proof is not empirical. This dispute is mixed, and each side presents it as purely its own kind.

Frontier Values dispute one: who may consent to a release that crosses borders? The strongest sovereignty position: a state with a functioning biosafety authority is the legitimate decision-maker for its territory, and requiring global agreement would make African disease burden hostage to votes cast elsewhere — a point African delegations made in 2018. The strongest community position: the 2022 decision already names free, prior and informed consent of communities, the affected parties are households rather than ministries, and a village that gains nothing directly from a trial is hosting a risk for a research programme. The strongest position that neither suffices: the affected unit is a mosquito population spanning several countries and no body has authority over it — the Cartagena expert group asks whether assessment should cover the entire area of potential spread, and does not answer. No measurement resolves this. The empirical part, how far a drive would actually spread, is real, unmeasured, and does not by itself decide who should be asked.

Frontier Values dispute two: is irreversibility a special category of harm? One position holds that it is: an act that cannot be undone removes the option value every other public-health decision retains, and should face a higher bar regardless of expected value. The other holds that malaria is also irreversible for the roughly 610,000 people it killed in 2024, that the status quo is not a neutral baseline, and that treating inaction as costless is a value judgement smuggled in as caution. No measurement resolves this either. It is a disagreement about weighing certainties against possibilities, and it is the real content of most exchanges that appear to be about ecology.

Frontier Values dispute three: whose programme is it? The critics’ framing is that a technology developed in European laboratories and funded from Seattle is being tested on West African villages, and that the funder’s presence in national scientific capacity is itself a form of pressure. The developers’ framing is that African institutions, regulators and investigators hold the decision rights and that refusing them the tool is the paternalism. One empirical hook: whether evaluation independent of the developer produces different answers is measurable, and untested here because there is nothing yet to evaluate.

Speculative An open question belonging to neither category yet: what a partial failure looks like. The modification-drive result shows a drive decaying into a resistant, non-driving transgene distributed through a wild population. No governance instrument has a vocabulary for that outcome — not a release, not a harm, not a success — and no monitoring programme is designed to detect it. Flagged speculative because it has been seen only in cages; noted because a category of outcome with no name is a category no risk assessment will contain.

Established The binding constraint is not the drive. It is permission, and permission is blocked on evidence that only permission can produce. Cage suppression is reproducible in two architectures and two species, and the number of open releases is zero. Every bottleneck below follows from that fact rather than being a separate engineering problem.

Frontier Detection at low allele frequency is the bottleneck nobody owns. A drive released at a few per cent must be tracked through the phase in which it is rare — exactly the phase in which a spread boundary would be violated and a recall would still be possible. That needs standing entomological surveillance across a region, genotyping throughput, and a sampling design powered for absence rather than presence. No published programme reports such a network, and the Cartagena expert group lists monitoring over the relevant scales as an open problem.

Frontier There is no demonstrated recall. Reversal proposals exist on paper — an immunising element, a second drive that overwrites the first, flooding with wild-type — and none has removed a drive from a wild population, because no drive has been in one. The adopted guidance states the problem in its own terms: a potential inability to halt the spread or reverse its action and effects. A countermeasure never exercised is a plan, not a capability.

Frontier Baseline ecological data are missing at the scale a trial would need. Detecting a food-web effect requires several seasons of pre-release measurement of endpoints chosen before the release rather than after, and the modelling review finds food-web and ecosystem-service impacts absent from the literature altogether, so there is not even an agreed list of what to measure. This is a cheap bottleneck by the standards of the others and it has not been cleared.

Frontier Strain fitness in the field was measured once and the measurement was unfavourable. That matters less for a self-sustaining drive, which need only enter the population once, than for the confined threshold-dependent designs every proposed first trial relies on — precisely the designs whose success depends on release numbers and mating competitiveness.

Established Regulatory capacity is a bottleneck with published numbers behind it. Most African states have no gene-drive-specific provisions; two have genome-editing instruments; three apply general biosafety law; nine report significant public opposition. Burkina Faso illustrates the coordination failure exactly: the biosafety agency had approved earlier trials, the environment ministry required impact studies, and the suspension came from a third body.

Frontier The definitional bottleneck is small, boring and genuinely blocking. Regulators cannot agree whether the regulated product is the released mosquito or the modified wild population that follows. On the first reading, releasing a few thousand insects is a minor event; on the second, the same act permanently modifies a transboundary population. A dossier cannot be written against a definition that does not exist.

Speculative Funding structure is a bottleneck that arrives later and has no obvious solution. A one-shot intervention with no recurring sale fits none of the instruments that pay for vector control, all built around repeat procurement of nets, insecticide and drugs. Nothing in the record settles who pays for fifty years of post-release monitoring on a product sold once, or never sold at all.

Frontier The dependency that gates everything else is a measured gene-flow map of the target population. Whether a release is local depends on how far alleles actually move in the Anopheles gambiae complex across seasons, which is a population-genomics measurement rather than a modelling assumption. The Cartagena expert group turns this into a legal question — whose protection goals frame the assessment — but the legal question is downstream of an empirical one answerable by sampling and sequencing, without any release.

Frontier Models must acquire the ecology they currently omit. The review of 117 modelling publications finds most of them deterministic, non-spatial and generic, with mating-related factors, species interactions, behavioural resistance, hybridisation and predator-prey relationships largely or wholly excluded. Until spatial, species-specific, stochastic models exist for a candidate release area, the confinement and ecological-safety claims both rest on tools their own reviewers say cannot carry them.

Frontier Resistance genetics needs wild genetic backgrounds, not colony backgrounds. The strength of the doublesex result is that the target does not tolerate functional change; the untested part is whether wild populations carry standing variation at or near that site that colonies do not. This is answerable from existing collections and sequence archives, and is the cheapest high-value experiment in this brief.

Established Trial infrastructure is a dependency on other people’s epidemiology. Any efficacy claim in humans needs the cluster-randomised machinery vector-control trials already use: defined clusters, enrolled cohorts, parasitaemia or incidence endpoints, sites characterised for years beforehand. It exists, it is expensive, and it is not built for an intervention that crosses cluster boundaries on its own.

Frontier Social-science evidence is a dependency the field has been told it treats as decoration. The governance analysis of the Burkina Faso suspension argues that engagement functioned as a feedback mechanism rather than something able to reshape research timelines; interviews with nineteen Ugandan stakeholders concluded that risk decision-making must be grounded in local context to be legitimate. Neither is a call for more communication; both are claims about where decision authority sits.

Frontier The decisive experiment is a cluster-randomised release of a confinable drive into a genetically isolated population — an island or an equivalently bounded deme — with ecological and epidemiological endpoints pre-specified before release and a monitored countermeasure held in reserve. It is decisive because it attacks the brief’s central fact: every efficacy, ecology and confinement claim in the field comes from a cage, a mesocosm or a model, and none can be discriminated without an open population. Randomisation converts an anecdote about one valley into an effect size; genetic isolation bounds the consequence of being wrong; pre-specification stops the ecological endpoint being chosen after the data arrive.

Established The design is not novel; its application here is. Cluster-randomised trials with entomological and parasitological endpoints are the standard apparatus of vector-control evaluation, and the World Health Organization’s guidance framework for testing genetically modified mosquitoes, second edition, sets out a phased pathway from contained work through cage demonstration and modelled scenarios to field release, describing the step for self-sustaining drives as a stepwise process that becomes a continuum of expanding releases. That phrase is the honest one: for a self-sustaining construct there is no discrete trial, only a release that keeps going.

Frontier The methodological difficulty is specific, and it is why this experiment has not been designed in detail anywhere public. Cluster randomisation assumes the intervention stays in its cluster. A drive is engineered not to. Either the trial uses a threshold-dependent or self-limiting construct, testing a different object from the one that would be deployed, or it uses the real thing and the control clusters are contaminated by design. This is a genuine trade and the brief does not resolve it.

Frontier Genetic isolation has to be measured, not assumed. The conservation literature challenges the island premise directly, arguing that the narrative of contained gene drive use on islands substitutes geography for gene flow. The precondition is therefore a prior, cheaper experiment: sequence the candidate population and its nearest neighbours across seasons and measure migration. If that shows appreciable gene flow, the decisive experiment cannot be run there and the field has learned something important without releasing anything.

Speculative The reserve countermeasure is the weakest part of the design. An immunising element or overwriting drive held ready is what would make the release conditionally reversible, and no such countermeasure has been exercised against a drive in a wild population. A protocol listing one asserts a capability nobody has demonstrated. The honest version either includes a prior demonstration of recall in a large enclosure, or states plainly that the release is one-way and asks for consent on that basis.

Frontier Three cheaper experiments would each move the assessment, and two require no release. Sequencing wild populations at and around the doublesex target, to test whether standing variation colonies lack would generate functional resistance. A multi-season baseline of the ecological endpoints a suppression trial would use. And a large-enclosure demonstration of a recall — the only one involving a drive outside containment, and the only one that turns a reversal plan into evidence.

Frontier A natural experiment is already running. Australia’s regulator has approved contained work on a rodent drive, published in 2022, with enclosure trials under semi-natural conditions, against modelling that about 250 modified mice could eradicate an island population of 200,000 in around twenty years. If the enclosure result diverges from that model, it is the first empirical check on the class of prediction the confinement debate rests on.

Speculative What would change this assessment fastest is not a success but a failure. A confined release in which the drive crossed its pre-registered boundary, or in which functional resistance appeared in a wild background, would settle two of the four factual disputes in a single season, in the direction current evidence does not anticipate. The field has arranged itself so that only a positive result is expected to be informative.

Established Nobody has scheduled it. No country has agreed to host a gene drive release, no funder has published a cost for the trial described here, the programme that came closest was terminated in August 2025 with no successor host announced, and the World Health Organization has no policy recommendation for gene drive mosquitoes. The horizon on the decisive experiment is therefore unscheduled rather than distant, and that is a governance statement rather than a technical one.

Established What has to be built is almost entirely monitoring and institutional capacity rather than biology. The constructs work in cages. What does not exist is the apparatus that would let anyone see what a construct is doing once it is outside one.

Frontier A regional surveillance network able to detect a drive allele below one per cent. Fixed trapping sites across and beyond the release area, sustained for decades rather than a grant cycle, with sample archiving, cold chain and sequencing capacity sited in the host country rather than the developer’s laboratory. The independence matters for the reason section 4 gives: the one proposal both developers and their critics endorse in some form is that evaluation be separable from the programme evaluated.

Frontier A pre-release ecological baseline, funded as infrastructure rather than as part of a trial. Several seasons of measurement on pre-specified endpoints is the only way the food-web argument becomes decidable, and the only element of this list that could be built today without resolving a governance question.

Speculative A held countermeasure, and a facility in which it has been exercised. As an engineering requirement this means a maintained stock, a release capability and at least one large-enclosure demonstration that a drive can be removed from a population it has entered. No such demonstration has been published.

Frontier Rearing and release capacity, where the self-limiting designs get expensive. A self-sustaining drive needs a small release; the confinable designs a first trial would use need repeated releases at a fraction of the wild population, which is a factory with sex-sorting and distribution attached. Moving a designed organism from a colony into open ground is treated in Designer Organisms; the editing capability sits in Genetic Engineering.

Established Nothing here requires a laboratory advance. That inverts the usual shape of a frontier assessment: the requirements are a monitoring network, a baseline dataset, a demonstrated off-switch and a decision procedure, three of the four institutional.

Established The nearest neighbours are the vector controls that already work, and they set the bar. Nets, indoor residual spraying, diagnostics and antimalarials are credited with 2.3 billion averted cases since 2000. Wolbachia-based incompatible-insect and replacement approaches are deployed or in trials with no engineered drive at all; the Hawaiian programme to protect native honeycreepers from avian malaria uses incompatible males rather than a drive and reached a draft environmental assessment in 2023. Self-limiting engineered strains and classical sterile-insect releases occupy the same niche with recurring cost.

Frontier Conservation applications are the second family, and they are where the consent question is sharpest. Invasive rodent suppression on islands, cane toads in Australia and grey squirrels in the United Kingdom have all been proposed. The overlap with deliberate ecological reconstruction is in Biodiversity Restoration and Synthetic Ecosystems, and the business of putting absent organisms back rather than taking present ones out is in De-Extinction Technologies.

Speculative Agricultural pest drives have the largest commercial pull and the least published evidence. They would put a self-propagating construct into a species with economic value to somebody, converting the consent problem into a trade question. Nothing in the fetched record reports a field-ready agricultural drive.

Frontier Three adjacent briefs own questions this one deliberately does not. Engineered organisms released into open ground for environmental purposes, and the laboratory-to-field gap there, is Environmental Biotechnology; engineered communities that will not establish are in Microbiome Engineering; the same containment logic pointed outward is Planetary Protection.

Established Four bodies claim a piece of this and none has authority over the thing itself. The Convention on Biological Diversity and its Cartagena Protocol supply risk-assessment guidance that is voluntary and decisions that are not legally binding. The World Health Organization supplies a testing framework and, as of late 2025, no policy recommendation on gene drive mosquitoes. The African Union Development Agency issued guidelines for risk analysis and for containment facilities in 2022. National biosafety authorities hold the only binding power, over territory the organism does not respect.

Established The two international frameworks pull in different directions and national regulators must satisfy both. A regulatory analysis puts it directly: the Convention’s guidance favours small, tightly confined experiments, mirroring the world of crops rather than mosquitoes, while the World Health Organization frames the question through public health — can this tool reduce disease? Biosafety authorities cite the first and health ministries follow the second, and the overlap obscures who is responsible and what evidence is required.

Established The Protocol’s central machinery does not fit. Advance informed agreement is built on an exporter notifying an importer before a shipment crosses a border. A drive crosses borders by mating: no exporter, no consignment, no moment of transit to regulate. The published commentary notes that gene drive movement cannot be managed by the traditional exporter-importer logic of biosafety protocols.

Established The largest research jurisdiction is outside the treaties. The United States is party to neither the Convention nor the Protocol, has no gene-drive-specific regulation, and splits jurisdiction among three agencies under a framework written for other purposes. The European Union would apply a 2001 directive designed around conventional modified organisms; Australia regulates through a dedicated office.

Frontier The regional layer is the one proposal that fits the organism, and it does not exist. The Cartagena expert group records that because some of these organisms may spread across jurisdictional boundaries, regional approaches facilitating multi-country oversight have been suggested. Suggested is the operative word: no standing regional authority can approve, condition or halt a transboundary release, and the African Union bodies nearest to it issue guidelines rather than permits.

Established Every party to this argument has a structural reason to shade it, and the brief labels them rather than discounting them. Target Malaria and allied consortia are (Gates-funded programmes) whose continued existence depends on a field trial eventually happening; the Outreach Network for Gene Drive Research and the GeneConvene Virtual Institute are (developer-side communication bodies); Revive & Restore and ISAAA are (organisations constituted to promote the technologies they assess). On the other side, the ETC Group, Friends of the Earth, Save Our Seeds and the Burkinabé coalitions are (campaigning organisations) whose mandate rests on the technology being stopped. National regulators are (interested parties too): an approval that goes wrong ends careers and a refusal never does. None of this makes anyone wrong. It predicts which errors each will make.

Frontier The Burkina Faso suspension is the best available natural experiment in how these institutions behave, and it flatters none of them. A biosafety agency had approved earlier trials; an environment ministry required impact studies; a research ministry suspended everything; authorisation for the final release came five days before it happened. Community engagement had been internationally praised, and the governance analysis concludes it lacked authority to redefine legitimacy thresholds or alter research timelines. A system in which approval, veto and legitimacy sit in three places, and the fastest step is the permit, is not one a developer can plan against or a community can rely on.

Established The consent problem is structural rather than procedural, and stating it plainly is not the same as amplifying it. A self-propagating construct released into a wild population will, if it works, reach territory whose inhabitants were not asked, because reaching them is the mechanism. No consent procedure at village, district or national level covers the area a successful drive would occupy. That is a fact about the technology, not a criticism of any programme, and it would remain true if every programme conducted exemplary engagement.

Established The instruments acknowledge it and do not resolve it. The 2022 decision under the Convention names free, prior and informed consent of indigenous peoples and local communities. The 2024 expert analysis asks whether risk assessment should address the protection goals of the host jurisdiction or of the entire area of potential spread, and leaves it open. Between a consent standard that applies to communities and a spread area that crosses states, there is no bridge in any current instrument.

Frontier Irreversibility is the second plain fact, and its weight rather than its existence is what is contested. The guidance adopted under the Protocol records a potential inability to halt spread or reverse action and effects, and nobody in the fetched literature disputes that description. What is disputed, as section 4 sets out, is whether irreversibility deserves a categorically higher bar than expected-value reasoning would give it — a values question on which this brief takes no position.

Frontier The benefit distribution is lopsided in both directions, which is why the argument is bitter. If a suppression drive works, those who gain most are among the poorest in the world and they gain something — children not dying of malaria — that no amount of procedural caution substitutes for. If it fails badly, the people who bear it are the same people, and the institutions that authorised it are, on the published capacity review, mostly under-resourced. Both halves are load-bearing and the debate usually drops one. A village hosting a release, meanwhile, gains no health benefit from it; standard human-subjects frameworks handle that badly because the subjects are not enrolled in anything.

Frontier The colonial framing is a serious argument and it is made from both sides. Critics hold that a technology developed in European laboratories and financed by a single American foundation, tested in West African villages, reproduces an old pattern; the Burkinabé coalitions framed their case around sovereignty and donor influence, and the governance analysis found those frames resonated more strongly nationally than the technical ones. Developers and many African scientists hold the reverse: that treating African regulators as incapable of deciding is the paternalism. A third position, from African vector biologists, is that the technology should proceed under nationally authorised, independently evaluated trials.

Speculative Dual-use is raised in the security literature and is thinner than the other concerns. Analyses note that a drive aimed at a crop, a pollinator or a keystone species would be hard to attribute, and that genome editing was flagged as an emerging dual-use concern in a 2016 United States intelligence assessment; section 15 says why the mechanism argues against it.

Frontier A gene drive would be the first technology whose deployed unit is a heritable change in a wild population rather than an artefact under someone’s control. Every prior environmental intervention — a dam, a pesticide, a vaccine campaign, a sterile-insect release — is sustained by continued human action and decays when it stops. A self-sustaining drive is sustained by the population it modified. The institutional consequence is that the decision to act and the decision to continue acting, which every regulatory system treats as separable, become one decision.

Frontier The precedent matters more than the mosquito. If a drive is released and works, the governance that permitted it becomes the template for every subsequent proposal to edit a wild population — agricultural pests, invasive vertebrates, species somebody else values. If something goes wrong, the same template gets blamed. A first case decided badly in either direction sets a rule for a class of interventions far wider than malaria control.

Speculative Eliminating a vector would be the first deliberate extirpation of a species achieved as public policy rather than as an accident. Smallpox and rinderpest were pathogens eradicated by immunity; this would be an insect removed from an ecosystem by design. Flagged speculative because no release has occurred, the cage results concern populations rather than species, and the Anopheles gambiae complex has several members of which a given drive would affect some and not others.

Handwave The claim that a successful drive would make subsequent releases routine is assertion. It is plausible, since institutional precedent usually lowers the next barrier, and it is equally possible that a first success would produce tighter rules, as happened in several nuclear and biomedical cases. Nothing in the fetched record bears on which. The step is flagged because it does real work in both the optimistic and the pessimistic versions of this subject.

Frontier The most durable fact here may be the demonstration that a technically ready intervention can be stopped by legitimacy alone. Burkina Faso suspended a programme that held valid permits, had no reported safety incident, and had conducted engagement its peers praised. Whatever one thinks of that outcome, it establishes that for this class of technology the binding constraint is consent rather than capability — which is why this brief spends more of its length on institutions than on biology.

These horizons track permissions, releases and evidence rather than laboratory capability, because laboratory capability has not been the binding constraint for several years:

• 10 yr: Frontier Contained and enclosure work continues and expands, including the Australian rodent enclosure trials and further caged mosquito results. Speculative A first confined field release of a self-limiting or threshold drive is possible in this window and less likely than it looked in 2023: the leading programme was terminated in August 2025, its insectary closed, no successor host country announced. Frontier The World Health Organization either issues a policy recommendation for gene drive mosquitoes or does not; it has none today. Frontier The Cartagena expert group reports to the next meeting of the parties, and either identifies gaps requiring binding measures or continues the voluntary guidance track. Speculative Regional oversight for transboundary releases is formally proposed by an African Union body; nothing on the record says it will be.
• 25 yr: Speculative A self-sustaining drive is released somewhere, or the technology is abandoned for malaria in favour of the Wolbachia and self-limiting approaches already deployed. Speculative The first epidemiological endpoint from a drive trial is reported, or it is not, and this brief is written on the assumption that it might not be. Handwave Conservation applications reach open release on an island; the confinement evidence base described in section 3 would have to be rebuilt from twelve modelling papers first.
• 50 yr: Handwave If drives are in use, the question becomes maintenance: whether resistance has accumulated, whether populations have rebounded, and whether anyone still funds the monitoring network section 8 says must run for decades. Speculative The governance precedent set by the first case, whichever way it went, is by then the operative rule for editing wild populations generally.
• 100 / 250+ yr: Handwave Any statement at this horizon is assertion. The two shapes usually offered — wild genomes routinely managed as infrastructure, or a permanent norm against self-propagating release established after an early failure — are both coherent and neither is supported by anything in this brief’s evidence. The long record of designing organisms to specification is in Designer Organisms; of attempting to run engineered ecosystems, Synthetic Ecosystems.

• Depends on Genetic Engineering, which owns nuclease-based editing as a general capability; this subject consumes it rather than producing it. The nearer sibling is Designer Organisms, which owns the record of organisms built to specification and released, including the non-driving mosquito strains and the cost structure of mass release. Nothing on this map is a scientific blocker. The blockers are below, and none is a result anyone is trying to produce in a laboratory.
• Requires (not on this map) Six things, none a laboratory result and none produced by any brief here. A measured gene-flow map for the target species complex, because whether a release is local is an empirical property of the population rather than of a coastline. A demonstrated recall, since reversal exists on paper and has never been exercised against a drive in the wild while the adopted guidance records a potential inability to halt spread or reverse effects. A surveillance network able to see a drive allele below one per cent, sustained for decades and sited independently of the developer. A consent procedure for a release that crosses a border, between a 2022 decision naming free, prior and informed consent of communities and a 2024 analysis leaving open whether assessment covers the entire area of potential spread. A standing regional authority able to approve or halt a transboundary release, such approaches being recorded as suggested and none existing. And a financing instrument for an intervention sold once or never.
• Enables If the confinement and gene-flow questions resolve favourably, the same machinery applies to invasive rodents on islands, to agricultural pests, and to any case where what is wanted is a change in a wild population rather than a product delivered to it. If they resolve unfavourably, what is enabled is a monitoring and baseline capability that Biodiversity Restoration would find useful regardless.
• Adjacent Synthetic Ecosystems for engineered ecological systems failing in ways their designers did not model; De-Extinction Technologies for the mirrored problem of putting organisms back; Environmental Biotechnology for the gap between laboratory and open-ground performance; Microbiome Engineering for engineered organisms that will not establish; and Planetary Protection for the same containment logic pointed outward.

Established “Gene drives have been shown to eliminate malaria mosquitoes.” They have been shown to eliminate caged colonies. Two cages of 600 insects and six of about 570 adults, under controlled temperature, humidity and light, with no immigration and no competing species. The developers’ own papers do not make the larger claim: the large-cage authors write that comprehensive resistance testing and environmental risk assessment will be needed ahead of field trials.

Established “Gene drive mosquitoes have already been released in Africa.” No gene drive has been released anywhere on Earth. What was released in Burkina Faso in 2019 and again in August 2025 were modified strains that do not drive — a sterile-male line and a male-bias line whose modifications are not self-propagating. Both critics and supporters occasionally blur this, in opposite directions and for opposite reasons.

Established “Burkina Faso stopped the project because something went wrong.” No safety incident or regulatory breach has been reported. The government cited national interests; an independent governance analysis lists transparency, divided expert assessment and ecological doubt. Reading the suspension as evidence of an accident is a misreading both camps find convenient.

Frontier “Resistance has been solved.” It has been avoided at one target. The doublesex result is genuinely strong and replicated. At other targets it is not solved at all: the modification drive in Anopheles coluzzii generated resistant alleles persisting in about 30 per cent of individuals, a functional mutation that blocked homing, and instability in the guide array itself. Generalising from the best target to the technology is the commonest error in the optimistic literature.

Frontier “A gene drive will wipe out mosquitoes and collapse ecosystems.” The direct trophic evidence does not support the second half, and the apparatus to check it does not exist. Predation studies show substitutable prey and negligible biomass, while the modelling review finds food-web and ecosystem-service impacts absent from the literature. No large direct effect has been demonstrated and no method capable of demonstrating a subtle one has been built. Both the alarming and the reassuring versions overreach.

Established “The United Nations has banned gene drives.” It has not, and it declined twice. A moratorium was pressed in 2018 by Bolivia, El Salvador, Grenada and Egypt with substantial civil-society backing; the decision that passed called for precaution and case-by-case assessment instead. In 2024 the parties to the Cartagena Protocol welcomed voluntary guidance and established an expert group, with no moratorium language; the conservation sector voted separately in October 2025 and rejected a moratorium motion.

Frontier “Local drives solve the consent problem.” They would, if they were local, which is the thing in dispute. The claim rests on twelve modelling papers a systematic review judges insufficiently tested for firm conclusions, on models that are mostly non-spatial, and on an island premise the conservation literature challenges directly. That is not a reason to reject confinable designs; it is a reason not to treat their confinement as established while arguing that everything else must be.

Speculative “Gene drives are a plausible biological weapon.” The governance literature raises it; the mechanism argues against it. Attribution would be difficult because effects appear years later, which is exactly why a drive is a poor instrument of coercion: slow, uncontrollable by whoever releases it, and no more confinable to an adversary’s territory than to anyone else’s.

Frontier “There is no alternative; malaria control is failing.” Partly true and stated too strongly. Conventional tools are credited with 2.3 billion averted cases since 2000, the largest measured effect anything in this field has produced, and they are losing ground in specific places under insecticide resistance and an invading urban vector. A tool that is eroding is not a tool that has failed, and the honest comparison is against a fully funded conventional programme rather than a neglected one.

Handwave “The disagreement is really about science, and better communication would fix it.” This is the claim this brief most wants to retire. Two of the four disputes in section 4 are about measurements nobody has taken and would be settled by taking them. Three others are about who may decide, what weight irreversibility carries, and whose programme this is, and no measurement touches any of them. Treating the second group as a communication failure is what produced a programme with internationally praised engagement being shut down by the state it was engaging in.