Self-Spreading Vaccines

Speculative If a vaccine could be released into a wild animal population and spread through it on its own — and if it stayed what it was released as, which no programme anywhere has measured for longer than 32 days — the prize is immunity in species that cannot be caught one at a time. Rabies in vampire bats, Lassa in Mastomys rodents and bat white-nose syndrome are the three cases the policy literature argues about, and in each the obstacle is not that a vaccine does not exist but that nobody can hold millions of wild animals long enough to inject them.

Established The magnitude is best stated as the gap between what has been measured once and what has never been tested at all. The published record contains exactly one field release: 76 rabbits inoculated and returned to a population of roughly 300 on a 34-hectare island 1 km off Menorca, with antibody later found in about half of the 71 animals nobody had touched, over a 32-day observation window. That is the entire empirical base for self-spread in the wild, it is 27 years old, and no authority anywhere has approved a second one. The largest public programme since, the United States Defense Advanced Research Projects Agency’s PREEMPT, ran for three and a half years, put up to $9.37 million into its biggest single award, and stated in writing that it was not funding release into the environment. So the upside here is bounded by a quantity no programme reports: nobody has measured what one of these agents does in a free population over a year, a decade, or a host generation.

Frontier The gains and the risks fall on different people, and that asymmetry is the part a legislator should hold onto. The benefit of suppressing Lassa is concentrated in rural Sierra Leone; the benefit of suppressing vampire-bat rabies among Latin American livestock holders, where a candidate bat vector has been found in 80–100% of animals across 25 Peruvian sites. The risk, once an agent is released, is carried by everyone downstream of the animals that move, in whatever country they move into, for as long as it persists — and the deepest pocket that has funded the work was a military agency whose stated interest was protecting deployed personnel.

Speculative For any of this to be worth having, one thing has to become true that is not true now: some body with jurisdiction has to be willing to say what evidence would justify an irreversible release. No such body has said it, and the bottleneck list in 5 is mostly about why.

Established A self-spreading vaccine is a vaccine designed to move between animals by itself, so that immunising a few individuals immunises a population. The idea is roughly forty years old and has been proposed under several names — self-disseminating, self-propagating, transmissible. What makes it a governance subject rather than a veterinary one is a single structural fact: the intervention is the release of a self-replicating agent into an open system, and once released it is not retrievable.

Established This brief deliberately contains no technical or operational content. It does not describe how such an agent would be designed, chosen, altered or made to persist, and it does not summarise the design sections of the papers it cites even where those are open-access. The reason is the reason the subject is contested: a substantial part of the opposition holds that the knowledge generated by doing this work is itself the hazard, independent of whether anything is released. A brief that reproduced that knowledge in order to explain the argument against reproducing it would be self-defeating. What follows is the policy debate, the evidence standard and the institutional gap.

Established The literature draws one distinction that carries most of the policy weight: transmissible versus transferable. A transmissible design spreads by infection from animal to animal and keeps going without further human input. A transferable design moves only by direct physical contact and stops when that contact stops. Nuismer and Bull set the distinction out in Nature Ecology & Evolution in 2020; Sandbrink, Watson, Hebbeler and Esvelt used it the following year in the same journal to argue for the transferable route and against the transmissible one. It is the closest thing the debate has to a dividing line between a reversible intervention and an irreversible one, and most public argument collapses the two.

Established The debate has a fixed cast. On one side, those who have published the case for proceeding: Scott Nuismer at Idaho, James Bull, Daniel Streicker at Glasgow, Michael Jarvis at Plymouth, and the PREEMPT teams. On the other, biosecurity analysts — Filippa Lentzos at King’s College London, R. Guy Reeves at the Max Planck Institute for Evolutionary Biology, Jonas Sandbrink, Kevin Esvelt — with conservation-policy organisations and, separately and unhelpfully, an anti-vaccine movement that has taken the subject up as a conspiracy narrative. Two Science Policy Forums, January 2022 and April 2024, are the formal statements of the disagreement.

Established One field release has happened, and it is the only empirical anchor either side has. In a trial reported in Vaccine in 2001, a recombinant myxoma virus carrying a rabbit haemorrhagic disease antigen was given to 76 wild rabbits on Isla del Aire, a 34-hectare island about 1 km off Menorca, Spain. All 76 developed antibody to both targets; about half of the 71 uninoculated contact animals did too, over 32 days. The paper states that no undesirable effects of the release were detected. The approach was not taken further. Everything else in the subject — every efficacy figure, every risk estimate — is modelling, laboratory work or argument.

Speculative The governance question is therefore not “does it work” but “who could say yes, and on what evidence.” A self-propagating release has no natural jurisdiction. It is not a medicine: there is no patient and no prescriber. It is not quite a conventional environmental release: the thing released reproduces and evolves. It crosses borders without a declaration. The institutions that would have to authorise it were built for other objects, and 10 is about which come closest and where each fails.

Established The Isla del Aire trial is a documented event and the commission’s description of it is accurate in substance. Torres and colleagues, in Vaccine 19(31):4536–4543 (14 August 2001), report the first field trial of a transmissible recombinant vaccine against myxomatosis and rabbit haemorrhagic disease. The work was “undertaken in an island of 34 Has containing a population of around 300 rabbits”; 76 rabbits were inoculated and all seroconverted against both targets; “the recombinant virus exhibited a limited horizontal transmission capacity, promoting seroconversion of around 50% of the uninoculated rabbit population”. A further 71 animals were tagged as contact, non-vaccinated controls. The observation period was 32 days. The island lies about 1 km off Menorca, in the Balearic Islands.

Frontier The date is the one part of the anchor the primary source does not carry. The paper is a 2001 publication whose abstract does not say when the animals were handled; secondary accounts disagree by a year, National Geographic placing it in 1999–2000 and the Max Planck summary in 2000. Nothing in the policy argument turns on which is right, and this brief uses 1999 while flagging the year as a secondary-source figure. The same caution applies to the transmission number: the paper says “around 50%”, one journalistic account says 56%, and the abstract is the figure to carry forward.

Established What the trial measured is narrow; what it did not measure is the entire governance question. It measured antibody in recaptured animals on one island, in one species, over 32 days. It did not measure what the agent did over a host generation, whether it stayed what it was released as, whether it reached the mainland, or what happened after the observers stopped looking. The paper reports no undesirable effects detected — a negative finding over a month on 34 hectares, which is a very different claim from safety.

Established The approach stopped for regulatory and funding reasons, not because it failed. On the account of Juan Bárcena, a doctoral student on the trial under José Manuel Sánchez-Vizcaíno, the European Medicines Agency asked for further work before considering marketing authorisation, allowed roughly two years, and the money did not arrive; his own view, reported by National Geographic, is that approval was unlikely anyway. The Max Planck summary puts it more bluntly: the Agency declined approval. The one time this question reached a regulator, the regulator did not say yes — and never had to state a standard it would have said yes to.

Established The largest public programme in the subject was military, ran three and a half years, and excluded release. DARPA announced PREEMPT — PREventing EMerging Pathogenic Threats — on 4 January 2018 under programme manager Jim Gimlett, a 3.5-year effort to address viral disease “at the source, animal reservoirs”. Five performer teams were announced on 19 February 2019 under Brad Ringeisen, led by Autonomous Therapeutics, UC Davis, Institut Pasteur, Montana State University and the Pirbright Institute. The UC Davis award was up to $9.37 million, with the University of Idaho (Nuismer), the University of Plymouth and its spinout The Vaccine Group (Michael Jarvis), the University of Glasgow, the University of Western Australia and partners in Sierra Leone including the government, the University of Makeni and Njala University. The targets were Lassa and Ebola.

Established DARPA’s own words on release are the single most quotable governance fact in the subject. Its February 2019 announcement states: “DARPA is not funding the release of PREEMPT interventions into the environment.” The programme page, which now records the programme as complete, states that “research is performed entirely in controlled laboratory facilities, including planned proof-of-concept demonstrations at the end of the program. If the program is successful, potential future field trials would take place under the auspices of other government stakeholders following all standard protocols for biosafety.” That sentence is a referral to nobody in particular. It names no agency, no statute and no standard, and assumes an authorising body downstream that, as 10 argues, does not exist.

Established The formal disagreement is on the record in Science, twice. In January 2022 a Policy Forum by Lentzos, Rybicki, Engelhard, Paterson, Sandholtz and Reeves, “Eroding norms over release of self-spreading viruses”, argued that releasing modified viruses able to spread between vertebrate hosts breaks a consensus holding since the mid-1970s, and that the work was proceeding without new evidence justifying the departure. The developers replied in the same journal with “Self-spreading vaccines: Base policy on evidence”. In April 2024 a second Policy Forum — Streicker and twenty-two co-authors, Developing transmissible vaccines for animal infections, Science 384(6693):275–277 — proposed “a series of commitments and strategies for vaccine development”. Neither was fetchable for this brief; the publisher refuses automated access, so both are cited by name and the 2024 abstract is quoted from the United States Geological Survey’s publication record.

Established The 2024 Policy Forum is notable less for what it proposes than for what it concedes. Its abstract states that generating transmissible vaccines “likely requires modifying viruses that would be intended to spread in nature, which raises concerns ranging from technical feasibility, to safety and security risks, to regulatory uncertainties”, and that the commitments are ones the authors “believe increase the likelihood that the potential risks of vaccine transmission are outweighed by benefits”. That is the field, in its strongest collective statement, describing its governance position as a belief about a likelihood.

Frontier The field has also published a substantial negative result about itself. Bull, Nuismer, Remien, Griffiths and Antia, in Expert Review of Vaccines 23(1):294–302 (2024) — four of the five among the technology’s principal proponents — argue that the safest class of design, one using a vector already endemic in the target population, should not be expected to expand its footprint of active infection at all. Containment, they write, “may limit vaccine efficacy, but it also promotes safety – the vaccine should not escape”, and high coverage would need repeated releases or release into a population lacking the vector, at which point “containment then becomes a concern”. The designs that are safest are the ones least likely to work, and the proponents say so.

Frontier The strongest field evidence for feasibility is ecological, not clinical. Griffiths and colleagues, in Nature Communications 11:5951 (2020), surveyed vampire bats at 25 Peruvian sites with evidence of rabies and found a candidate bat herpesvirus in 80–100% of animals, with host specificity within neotropical bats. The same paper records a regulatory fact that cuts the other way: “WHO guidelines for recombinant dog rabies vaccines discourage those with potential to be shed and spread infectiously”. The nearest existing international guidance discourages exactly the property the technology depends on.

Established The natural history of the one virus used in 1999 is a standing argument for caution, and it is measured rather than speculative. Myxoma virus was deliberately released into European rabbit populations in Australia and Europe in the early 1950s and is, per Águeda-Pinto and colleagues in Viruses 11:530 (2019), “the best-documented field example of host–virus coevolution, following a cross-species transmission”. Their paper characterises a naturally recombinant myxoma virus found in Iberian hares in Spain, a genus that was not the original host. Deliberate release, cross-species jump and recombination in the wild are this lineage’s documented history, not hypotheticals.

Established As of December 2025 the field had produced no candidate past contained work. The Third World Network’s Biotechnology & Biosafety Series paper of 3 December 2025 examines Lassa in Mastomys rodents, rabies in bats and white-nose syndrome in North American bats, and records that despite claims of being near-product since the 1980s, no candidate has “progressed beyond models and contained experiments”. It is an advocacy document; that particular claim is consistent with everything else here.

Established The position, stated plainly: one field release in 1999, one regulator that did not approve it, one military programme that ran 3.5 years and released nothing, two competing Policy Forums, a modelling literature, and no safety evidence base at population scale anywhere. Every claim about controllability, durability or efficacy at scale rests on models calibrated to parameters the modellers describe as unknown. That absence is the finding, not a gap in this brief.

Frontier First: is irreversibility a category difference or a matter of degree? Proponents say degree — a design confined to a vector already endemic in the target population is, on their own modelling, unlikely to expand, and an island release is bounded by water. Opponents say category — an agent that reproduces has no recall procedure and no way to be un-decided. Both positions are coherent and the evidence that would separate them does not exist, because it would consist of observing a released agent for longer than anyone ever has.

Speculative Second: does the wildlife-reservoir premise survive contact with the surveillance literature? The case assumes the reservoir and the pathogen that matters can be identified in advance. Holmes, Rambaut and Andersen argued in Nature 558:180–182 (2018) that “no amount of DNA sequencing can tell us when or where the next virus outbreak will appear”, and that money is better spent on surveillance than prediction. If they are right, the technology is useful only for the small set of known, geographically stable reservoir-pathogen pairs — Lassa, bat rabies, a handful more — and not as general pandemic insurance. That is a much smaller prize than the framing implies, and the two literatures rarely address each other.

Frontier Third: is the research separable from the release? Sandbrink’s argument, in the Bulletin of the Atomic Scientists in June 2022, is that work on making a virus spread better generates knowledge that transfers to more dangerous organisms, so the hazard begins at the bench. The counter is that every branch of virology generates such knowledge, that singling out this application is arbitrary, and that prohibition here would not remove the capability from the world. This is the crux of the dual-use half of the debate and it is not empirically resolvable: it is a judgement about the marginal effect of one programme on a diffuse capability.

Speculative Fourth: is consent meaningful at the level at which this operates? A release affects people who are not patients, in territories whose governments may not have been asked, via animals that do not observe borders. The nearest machinery is the free, prior and informed consent of indigenous peoples and local communities written into the Convention on Biological Diversity’s gene-drive decisions. Whether it can bear the weight of a transboundary and permanent decision is open, and nobody has tested it on this class of object.

Frontier Fifth: can a competent authority be constructed at all? Gene drives are the nearest analogue and their history is instructive: Convention on Biological Diversity decision 14/19 (2018) and decision 15/31 (2022), voluntary risk-assessment guidance welcomed at the 2024 meeting of the parties to the Cartagena Protocol, and a 2025 International Union for Conservation of Nature congress that adopted a first synthetic biology policy while declining a moratorium on genetic engineering in wild ecosystems. Eight years of the most concentrated international attention any comparable technology has received produced guidance, not authority.

Speculative Sixth: what does the absence of an evidence base do to the argument itself? The tempting symmetry — no evidence of harm, no evidence of benefit — is false. The benefit case can be tested incrementally and reversibly; the harm case cannot, because the harm scenario is instantiated by the same act that would test the benefit. Whether that asymmetry should be decisive is the question the whole debate actually turns on, and it is about decision rules under irreversibility rather than about biology.

Established The binding bottleneck is not laboratory capability. It is that no instrument exists for authorising an irreversible environmental release of a self-replicating engineered agent. Every framework that could plausibly apply rests on two assumptions a self-propagating agent violates: that the thing authorised can be contained, and that the authorisation can be withdrawn. The December 2025 Third World Network paper puts it exactly — existing frameworks presume “containment and reversibility – assumptions incompatible with self-propagating agents”. That is an advocacy source making a structural claim, and the structural claim holds independent of the advocacy.

Established Second: there is no post-release monitoring capability anywhere at the duration the question requires. The only field trial observed for 32 days. To learn whether an agent remains what it was released as, reaches non-target species, or persists, somebody would have to sample a wild population repeatedly for years across a landscape, with an assay and a data infrastructure maintained by an institution that outlives the grant. Nothing of that shape exists. An intervention whose central risk unfolds over decades is paired with a monitoring horizon of one month.

Frontier Third: attribution. If an engineered agent were found circulating in wildlife, no routine system would identify it as engineered, trace it to a release, or distinguish an authorised release from an unauthorised one. Without attribution there is no liability, no compliance verification under any arms-control instrument, and no way to answer the accusation — which the disinformation record shows will be made regardless — that an outbreak was somebody’s release. The detection half of this is treated in Pathogen-Agnostic Surveillance; attribution is a distinct and less-funded problem.

Established Fourth: the evidence a decision-maker would need cannot be generated without making the decision. A contained study answers questions about contained conditions. The questions that make the policy hard — what happens over generations, across a landscape, among other viruses and other species — are questions about open systems, and the only instrument that produces data about open systems is the release. The experiment and the risk are the same act.

Frontier Fifth: funding structure, which is why the subject keeps stalling at the same point. The 1999 trial ended when a regulator asked for more work and the money did not follow. PREEMPT ran 3.5 years and stopped at the laboratory boundary. Money arrives for demonstration and departs before the long, expensive evidence-generating phase a regulator would need. Nothing in the funding landscape is shaped like a twenty-year ecological safety programme, and a safety case that is inherently a long-duration field question cannot be financed on a three-year grant cycle.

Speculative Sixth: public legitimacy, already damaged before any deployment exists. By 2021–2022 the subject had been absorbed into anti-vaccine narratives: Children’s Health Defense published a piece in June 2021 framing self-spreading vaccines as the end of informed consent, and Coda Story documented in July 2022 how the research was misrepresented as a covert human vaccination plan, partly via a fabricated British government paper. Nuismer told that reporting he had “never heard of anyone working on transmittable vaccines for humans”. The policy consequence is real regardless of the falsity of the claims: any authority contemplating a release inherits a harder consent problem, not an easier one.

Established The research this subject most depends on is not virology. It is ecology, law and social science, and all three are underfunded relative to the biology. The papers carrying the technical case are numerous; the papers that would tell a regulator what to require are few, and mostly position pieces rather than studies.

Frontier It depends on reservoir ecology good enough to predict the consequences of suppressing a pathogen in a wild population. Andrew Peters, president of the Wildlife Disease Association, told National Geographic that “our understanding of infectious disease dynamics in wildlife remain for the most part too simple to meaningfully predict outcomes”, and the same reporting raised the possibility of rodent population increases following Lassa suppression. That is a named expert stating the predictive science is not there — a research dependency, not a rhetorical objection.

Frontier It depends on a measured record of what happens to an engineered agent in a free population over time, and no such record exists for any agent of this class. A 2019 Evolutionary Applications paper on the evolution and containment of transmissible recombinant vector vaccines treats the problem theoretically: selection can favour variants that lose the protective element, which is at once a loss of efficacy and, on the authors’ framing, a route to limiting persistence. Whether real populations behave that way has never been asked of a real release. This brief deliberately does not discuss how such loss occurs or how it might be influenced; the governance point is that the rate is unmeasured and a regulator would need it.

Established It depends on legal scholarship that does not yet exist in usable form. There is a substantial literature on transboundary harm, on advance informed agreement, and on liability and redress. There is very little on an agent that crosses a border by reproducing rather than by being moved, where no shipment, exporter or importer exists to attach an obligation to. The Cartagena Protocol’s machinery is built around notification before a transfer; a self-propagating agent has no transfer event.

Frontier It depends on dual-use risk assessment methodology that can compare research programmes rather than merely flag them. Sandbrink’s 2022 proposal is exactly this: comparative risk-benefit assessment before funding, an explicit commitment never to optimise transmissibility, a preference for the lowest-risk route where benefits are comparable, and an Asilomar-style multidisciplinary process. None of it is institutionalised for civilian wildlife-health research anywhere.

Established The commission for this brief proposed a decisive experiment — a contained island release with pre-registered endpoints — and asked whether it is the right experiment and whether it could be governed. The answer to the first is partly, and to the second is no.

Established The first thing to say about a contained island release is that it has already been done. Isla del Aire was 34 hectares, about 1 km offshore, held roughly 300 rabbits, and the trial ran 32 days with 76 inoculated and 71 contact animals. Pre-registration and modern endpoints would improve it; they would not change what it is, a short observation of one species in a bounded system. The 1999 trial was not ambiguous about its own result. It was uninformative about the questions that make the policy hard.

Frontier The design defeats its own purpose, and this is the central point. An island is chosen because it limits spread; the governance question is what happens when spread is not limited. A trial that succeeds in containing the agent tells you about containment on that island, and a trial that fails has produced the harm rather than measured the risk. There is no version of this experiment in which a good outcome is informative about the uncontained case. The proponents’ 2024 modelling makes the same shape of point from the other direction: a vector-endemic release is expected not to expand, and release into a naive population is where “containment then becomes a concern”.

Frontier The endpoints that matter are longer than any trial is likely to run. Persistence, stability in the wild, non-target effects and evolutionary trajectory are multi-generation questions. A pre-registered protocol could specify them honestly and then fail to observe them, the normal fate of long-horizon ecological endpoints attached to short-horizon funding. A trial whose primary endpoint is seroconversion at 30 days and whose safety endpoints fall at 20 years is two studies, and only the first gets funded.

Established On governance: there is at present no body that could authorise it. Such a trial on Spanish territory would engage European Union deliberate-release law and, for any product ambition, the European Medicines Agency — the combination that stopped the last one without stating what would have sufficed. Elsewhere it would engage a national biosafety regulator whose statute assumes containment, the Cartagena Protocol if material crossed a border, the Convention on Biological Diversity’s non-binding precautionary language, and, if the host ranges across a frontier, a neighbouring state with standing to object and no forum to object in. No single authority holds the file.

Frontier So the decisive result this brief would change its assessment on is not an experiment. It is a policy result: a named competent authority accepting jurisdiction over an intentional, self-propagating environmental release and publishing, in advance, the evidence standard it would apply. Not a guidance document, not voluntary commitments by developers, not a funder’s internal review board — a body with legal power to say no, stating in public what would make it say yes. That single document would convert an unanswerable question into a research programme with a specification, and its absence is why every other experiment here is premature.

Speculative If such a standard were published, the experiment it would demand is a staged, monitored release with a pre-committed multi-decade observation programme and a funded monitoring institution — and the honest expectation is that the monitoring, not the trial, is what would prevent it. The obstacle is not timid regulators; it is that the evidence standard an irreversible act deserves is expensive on a timescale nobody funds.

Established Two smaller results would also shift the assessment, and neither requires a release. First, a published comparative risk-benefit assessment of a specific candidate programme against its non-transmissible alternatives, of the kind Sandbrink proposed in 2022, conducted by a body that is not the programme’s funder. Second, a documented instance of a national regulator refusing or approving a transmissible-vaccine application on stated grounds — even a refusal would be informative, because the 1999 case produced a non-decision and the subject has had no regulatory reasoning to argue with since.

Handwave What would not settle anything, despite being frequently proposed: more modelling. The models here are calibrated to parameters their own authors describe as unknown. Additional output built on the same unmeasured inputs raises confidence without raising evidence, and it is the most common way this literature grows while its evidence base does not.

Frontier And one natural experiment is already running, unplanned. The myxoma lineage released into European rabbit populations in the 1950s has been followed for seven decades, coevolved with its host, and produced a naturally recombinant virus in a different genus. Nobody designed it as a safety study and the counterfactual is unavailable; it is nonetheless the longest observation the world has of a deliberately released virus in a wild mammal population, and why the reversion and host-jump concerns are treated as serious rather than theoretical.

Established The engineering this subject requires is institutional; the technical questions are the ones this document does not cover, for the reason given in 2. First, a competent authority with jurisdiction over intentional self-propagating release and a statutory basis for saying no. The nearest candidates are national biosafety regulators operating under statutes written for contained use and deliberate release of plants and, latterly, insects. Australia’s Office of the Gene Technology Regulator and the European Union’s deliberate-release directive are the most developed; the United States, which funded most of the work, splits the framework across three agencies, has no provision specific to self-propagating agents, and is not a party to the Convention on Biological Diversity or its Cartagena Protocol.

Frontier Second, a transboundary procedure that works without a transfer event. Advance informed agreement attaches to a movement of living modified organisms between parties; an agent that arrives in a neighbouring state inside a bat has no exporter and no notification. Something would have to be invented — a prior-consultation duty triggered by the range of the host species, with a forum in which an objecting state has standing. Nothing of the kind exists.

Frontier Third, a liability and redress regime proportionate to permanence. The Nagoya–Kuala Lumpur Supplementary Protocol addresses liability and redress for living modified organisms, and the December 2025 Third World Network analysis argues its provisions, like the rest of the framework, presume reversibility. An adequate regime would have to price an act that cannot be undone, across jurisdictions, over a horizon longer than the institutions involved.

Established Fourth, a monitoring institution with a mandate longer than a grant. This is the concrete, buildable piece: a standing capability to sample wild populations, detect an engineered agent, distinguish it from wild relatives, and publish, for decades. The nearest existing capability is genomic pathogen surveillance built for human health, discussed in Pathogen-Agnostic Surveillance, which is oriented at the human interface and is not designed to answer “is this ours”.

Speculative Fifth, an international norm-setting step, which is what the 2022 Policy Forum was really asking for. Lentzos and Reeves argue that a consensus against releasing self-spreading modified viruses has held since the mid-1970s and is eroding by default rather than by decision. Whatever one thinks of the substance, the process claim is hard to dispute: no international body has deliberated the question and reached a conclusion, so the norm is being changed by the accumulation of funded projects. The decision machinery for technologies of this class is the subject of Existential Risk Governance.

Established Sixth, an oversight body for the research phase that is not the funder. PREEMPT’s ethics advisers were appointed by DARPA and advised DARPA. Better than nothing, and structurally the wrong shape: the body assessing whether a dual-use programme should proceed should not be the body whose programme it is. Research-ethics boards at performing institutions review human and animal subjects; none has a mandate over the biosecurity consequences of a programme’s existence.

Frontier Seventh, and least discussed: a decision rule for the case where the evidence cannot be obtained. Most regulatory design assumes uncertainty shrinks with study. Here the residual uncertainty is irreducible short of the act itself, so an authority would need an explicit published rule — a precautionary default, a staged-consent procedure, something — rather than discovering its rule in the first case it faces. The closest working regime for a one-way contamination decision taken under irreducible uncertainty is Planetary Protection.

Established Gene drives are the nearest governance analogue and the most useful comparison a legislator can make. Both deliberately release a self-propagating genetic intervention into a wild population; both raise transboundary and consent problems; both have attracted moratorium proposals. The difference is attention: gene drives got a United States National Academies report in 2016, Convention on Biological Diversity decisions in 2018 and 2022, voluntary risk-assessment guidance in 2024 and sustained engagement by the Australian regulator. Self-spreading vaccines got almost none. The less-scrutinised technology is the one with an actual field release behind it.

Established The second analogue is the 2018 controversy over DARPA’s Insect Allies programme. It proposed dispersing genetically modifying agents through insect vectors in agriculture, and drew a Science Policy Forum from Reeves and colleagues arguing it looked more like a delivery system than agricultural research — the closest precedent for the concern that defence funding of environmental genetic modification generates suspicion outrunning a programme’s scope. The commission for this brief named a 2018 Science policy forum on self-spreading vaccines; the research could not find one. The Policy Forums on this subject are January 2022 and April 2024; the 2018 items are the Insect Allies forum and a review of transmissible vaccines by Bull, Smithson and Nuismer in Trends in Microbiology. The correction is noted here rather than buried.

Established The third is wildlife immunocontraception: the same idea aimed at fertility rather than disease. Australian work in the 1990s and 2000s explored virally vectored fertility control for introduced rodents and rabbits; a 2005 Vaccine paper by Smith and colleagues, including Alec Redwood, examined species specificity in that system. It produced no deployed product. The line matters because a self-spreading agent aimed at reproduction sits technically adjacent to one aimed at immunity, which is the source of a large part of the biosecurity concern.

Established The fourth is conventional wildlife vaccination, the benchmark any proposal must beat. Oral rabies baiting has been used at continental scale in Europe and North America. It is expensive, needs repeating, and fails where animals cannot be reached — the gap the self-spreading proposal aims at. An honest assessment compares against baiting and improved surveillance, not against doing nothing.

Frontier The fifth is the human-vaccine frontier, where the public confusion lives and where nobody is working. Deliberate self-spread in humans is incompatible with informed consent as every research-ethics framework understands it, and no programme in the published record proposes it. The adjacent legitimate work — broad-coverage vaccines that still require administration — is covered in Universal Vaccines, and the distinction between a vaccine that is broad and one that is contagious is what the public discussion most often loses.

Established The sixth is environmental release of engineered organisms generally. The regulatory and ecological questions that apply to engineered microbes in soil and water apply here with the added feature of an animal host that moves; Environmental Biotechnology covers that regime, and the immunological side is in Immune Engineering.

Established No international instrument specifically governs the intentional release of a self-spreading engineered agent; the Convention on Biological Diversity and its Cartagena Protocol are the closest fit and are not binding where it matters. The Protocol governs transboundary movement of living modified organisms through advance informed agreement, which presumes a movement and a mover. Convention decisions have reached self-propagating interventions only by the gene-drive route: decision 14/19 (2018) called for precaution and for release to be considered only where science-based risk assessments have been carried out; decision 15/31 (2022) reaffirmed case-by-case assessment, the free, prior and informed consent of indigenous peoples and local communities, and consideration of non-target effects. Convention decisions are not legally binding, and the United States — which funded the largest programme in this subject — is party to neither instrument.

Established The Biological Weapons Convention prohibits development for hostile purposes and says nothing about peaceful release. A self-spreading agent developed for public health is, in capability terms, close to something prohibited if the purpose were different, and purpose is not verifiable from the object. Lentzos and Reeves note the Convention is approaching fifty years old with weak compliance-assessment machinery, and that military funding of this class of work “could cause countries to doubt one another’s intentions”. That cost accrues even if no release ever happens.

Speculative The Environmental Modification Convention is sometimes invoked and is a poor fit, because it too turns on hostile intent. Nothing in the treaty architecture regulates a benign-intent, permanent, transboundary biological modification, which is the object in question.

Established National regulators are the only bodies with actual power, and their statutes were written for something else. The European Union’s deliberate-release directive and the European Medicines Agency between them stopped the 1999 line without producing a standard. Australia’s Office of the Gene Technology Regulator, the most experienced regulator of engineered organisms intended for wild populations, has so far authorised contained research rather than release. The United States splits responsibility across three agencies with no provision for self-propagating agents. In each case the statute assumes an authorisation that can be withdrawn.

Established The World Health Organization has the nearest thing to a substantive position, and it points away from the technology. The Griffiths 2020 paper records that “WHO guidelines for recombinant dog rabies vaccines discourage those with potential to be shed and spread infectiously”. That is a technical guideline for one product class rather than a policy on self-spreading vaccines, but it is the only extant international guidance that speaks to the defining property, and it discourages it.

Frontier Conservation institutions have begun to engage and have declined to prohibit. At the 2025 International Union for Conservation of Nature congress the motion that became the organisation’s first global policy on synthetic biology was adopted, while a separate motion calling for a moratorium on genetic engineering of wild species and ecosystems was not approved. That outcome is reported here from a secondary account and from the campaign that sponsored the failed motion, both interested parties. The direction of travel is clear: policy over prohibition.

Established Funders have filled the vacuum with self-regulation, and self-regulation is what it is. DARPA constrained its own programme to contained work, appointed its own ethics advisers and required its own performers to demonstrate controllability to it; the 2024 Science Policy Forum proposes commitments by developers. Both are better than their absence. Neither survives a change of funder, a change of jurisdiction, or a developer who declines to sign up, and neither can bind a state.

Speculative What a minimally adequate institution would look like can be stated even though it does not exist: a body with statutory power over intentional self-propagating release; a published evidence standard settled before any application; a transboundary consultation duty triggered by host range; a funded monitoring mandate outlasting the political cycle; a liability regime that prices permanence; and a decision rule for the irreducible-uncertainty case. Every element of that list is absent today, which is why the honest answer to “could a trial be governed” is no.

Established Consent is the ethical centre and it fails in a way that has no clean analogue. The animals cannot consent; the people affected are not patients and are not asked; the states affected may not be parties to the decision. Research ethics governs the relationship between an investigator and a subject; here there is no subject and many affected parties, a structure research-ethics boards were not built for and mostly cannot review.

Established Irreversibility changes the moral arithmetic rather than merely raising the stakes. A decision that can be revisited distributes responsibility across time; one that cannot concentrates it in the moment of the act and binds people who were not consulted because they do not yet exist. That is why the precautionary framing recurs in every critical document here. Stated soberly: a self-propagating release cannot be recalled, crosses borders without consent, and continues to evolve after the decision that authorised it has been forgotten.

Frontier The distributive question is real and it is not symmetric. The benefits of suppressing Lassa or bat rabies fall largely on people in low- and middle-income countries who bear those diseases now; the risks fall on everyone downstream, including those same people. A refusal to proceed is not neutral — it leaves a measured burden of disease in place. The precautionary side owes an account of that burden; the permissive side owes an account of who carries the residual risk and on whose authority.

Established The dual-use character is not hypothetical, and the strongest version of the concern is not about the vaccine. It is that the research programme itself — the knowledge, the trained people, the normalisation — is the hazard. Sandbrink, Watson, Hebbeler and Esvelt argued in 2021 that the safety and security risks of advancing transmissible vaccines outweigh the benefits and that the transferable route should be preferred. That is four biosecurity researchers, including one of those most associated with engineered self-propagating systems, saying the work should not proceed in its transmissible form. It is the strongest form of the case against and deserves reading in the original.

Established Interested parties, flagged hard, in all directions. The proponents have a line of work to defend: Nuismer, Bull and their collaborators have built careers on this technology, and their papers are the source of most of the efficacy modelling. Jarvis founded The Vaccine Group, a company in the PREEMPT consortium, and Plymouth reported a commercial holder with a 19.2% stake in it (company and university announcement). DARPA is a defence agency whose stated interest was force protection, and the $9.37 million figure is (the funding institution’s own announcement). On the other side, the Third World Network is an advocacy organisation opposed to environmental release of engineered organisms as a matter of standing policy, and the December 2025 paper is (an advocacy publication); the campaign behind the failed 2025 moratorium motion is run by an organisation whose purpose is to secure that moratorium (a campaigning organisation’s own account). Children’s Health Defense is an anti-vaccine advocacy group (an advocacy organisation with a prior position on all vaccines). The Max Planck summary of the 2022 Policy Forum is (the authors’ own institution). None of this makes anybody wrong. All of it should be visible when weighing what each says.

Established The durable civilizational question is about a norm, not a product. Since the mid-1970s the working consensus in biology has been that self-replicating engineered organisms stay inside containment. That consensus was never a treaty; it was a professional norm reinforced by funders, journals and institutional review. The 2022 Policy Forum’s title — eroding norms — is the argument in three words: norms of that kind do not fall, they thin, and they thin through funded projects that each seem reasonable.

Speculative If that norm goes, it goes for more than vaccines. The same structure — release a self-propagating agent to achieve a public good in an open system — describes gene drives for disease vectors, engineered organisms for ecological restoration, and field genetic modification in agriculture. A precedent set in the least-scrutinised of these transfers to the others without anyone deciding that it should. The first authorised release of a self-spreading engineered agent will set the evidentiary bar for everything after it, however casually it is set.

Frontier The verification problem it creates is permanent and probably worse than the release problem. A world in which self-spreading engineered agents circulate legitimately is one in which an unexplained outbreak has an additional plausible explanation and states have an additional reason to suspect one another. The Biological Weapons Convention has no verification protocol. Lentzos and Reeves make this point about military funding; it generalises to civilian funding once the capability is normal. The cost is paid in attribution and trust whether or not anything is ever released.

Speculative There is a real prize on the other side of the ledger and it should not be understated. If wildlife reservoirs are a material source of human pandemic risk, and if individually vaccinating wild animals is infeasible at the scales involved, a method of reaching those populations addresses a genuine and currently unaddressed pathway. The zoonotic disease burden is not disputed; what remains unevidenced is the claim that this technology can reduce it. A civilization that declines this line permanently is accepting the reservoir risk permanently, and should say so rather than pretend the refusal is free.

Frontier The subject is also a test case for whether a research community can decline a capability deliberately. The field has gone quiet: PREEMPT is complete, no successor of comparable size is visible in the public record, no candidate had left contained work as of December 2025, and the proponents’ most recent substantive paper argues the safest designs will not spread. That quiet is not a decision. It is a funding cycle ending and a technical difficulty biting, and either could reverse. A moratorium is on the record and survives a change of funder; a lull is not and does not.

These horizons track when a specific, checkable governance or evidentiary fact would have to be true, not when a product might appear. The dates are about decisions and documents, because that is what the record in this subject actually consists of.

• 10 yr: Frontier The realistic ceiling for the decade is a published evidence standard from some national regulator, most plausibly in a jurisdiction with an existing gene-technology regulator, triggered by a gene-drive application rather than by a vaccine one. Speculative A second field release of a self-spreading vaccine within ten years would require an application to be filed with a regulator that has not said what it would require; nothing in the public record suggests one is being prepared. Frontier Expect continued modelling, continued contained work, and continued absence of the monitoring infrastructure described in 5.
• 25 yr: Speculative A quarter-century is enough for an international instrument to be negotiated if a triggering event occurs; absent one, the gene-drive precedent suggests guidance rather than jurisdiction. Speculative If any authorised release happens on this horizon it will most likely be a gene drive against a disease vector, and the vaccine case will inherit whatever standard that process produces. Frontier The relevant comparison brief is Existential Risk Governance, whose central question — how a deliberative body acquires jurisdiction over a decision nobody currently owns — is this subject’s binding constraint too.
• 50 yr: Speculative The question at fifty years is whether the mid-1970s containment norm survives in recognisable form. Speculative If it does not, the governing question shifts from whether to release to how releases are registered, attributed and monitored. Handwave Any claim about disease burden avoided on this horizon extrapolates from models whose inputs are unmeasured.
• 100 / 250+ yr: Handwave At this range nothing specific to this technology is forecastable, and the honest statement is that a self-propagating agent released in the twenty-first century would still be evolving. Handwave The myxoma lineage released in the 1950s is at seventy-odd years and still producing recombinants in new host genera; that is the only observation the world has at even a fraction of this horizon, and it is one lineage, one virus, one accident of history. Speculative What is foreseeable is the institutional residue: whichever body first takes jurisdiction will still be the precedent, in the way that Planetary Protection still runs on decisions taken in the 1960s.

• Depends on On this map it depends least on biology and most on the governance machinery catalogued in Existential Risk Governance, and on the detection and attribution capability discussed in Pathogen-Agnostic Surveillance, which is the only existing technology that could tell a state whether an engineered agent is circulating in its wildlife. It depends on the environmental-release regulatory tradition covered in Environmental Biotechnology. It does not depend on any unsolved result in Universal Vaccines or Immune Engineering: the immunological problem here is ordinary, and stating that plainly is part of the assessment.
• Requires (not on this map) Five things, none of which is a laboratory result. A body with statutory power to authorise or refuse an intentional self-propagating release, which no state has designated. A duty to consult neighbouring states triggered by the range of the host species rather than by a shipment, because the Cartagena Protocol’s advance informed agreement attaches to a movement and a self-propagating agent has no mover. A liability and redress regime capable of pricing an act that cannot be undone, where the existing supplementary protocol presumes reversibility. A measured record of what happens to any engineered self-spreading agent in a free population over more than the 32 days ever observed — evidence, not engineering. And funding for monitoring on a horizon longer than the grant cycles that have ended every programme in this subject so far.
• Enables If the authority existed it would enable not this technology but the whole class: gene drives, engineered organisms for ecological restoration and any other deliberate release of a self-propagating construct would inherit a published evidence standard instead of negotiating one case by case — the strongest argument for building the institution regardless of what one thinks of self-spreading vaccines. On the technology’s own terms, success would enable population-scale immunity in reservoirs that cannot be individually reached.
• Adjacent Adjacent to gene drives, which share the structure and have received the governance attention this subject has not. Adjacent to wildlife immunocontraception, its closest historical relative and the source of a large share of the dual-use concern. Adjacent to Planetary Protection, the only working regime the world has for a one-way contamination decision taken under irreducible uncertainty, and the best available model for what a competent authority here might look like. Adjacent to Universal Vaccines in public perception and almost nowhere else — the confusion between a vaccine that is broad and a vaccine that is contagious does real damage to both subjects.

Established “Self-spreading vaccines for humans are being developed.” They are not, and no document in the published record proposes it. Every programme named here targets animal reservoirs. Nuismer stated in 2022 that he had “never heard of anyone working on transmittable vaccines for humans”, and Bull noted the DARPA Lassa work targeted a rodent virus that does not infect people. The claim circulates widely regardless. The consent objection is a real objection to the animal case and does not need the human fabrication; the fabrication has made the real objection harder to state.

Established “It has never been tried.” It has, once, and the trial is in the peer-reviewed literature. The 1999 release on Isla del Aire is reported in Vaccine in 2001 with numbers. Anyone arguing either side without that result in hand is arguing about a hypothetical that is not hypothetical.

Established “The 1999 trial proved it is safe.” It showed no detected harm over 32 days on 34 hectares. The paper says exactly that and claims no more. Treating a one-month negative observation in a bounded system as evidence of safety for an uncontained release is the commonest overreach in the subject’s popular coverage, and it is not one the original authors made.

Established “DARPA was going to release engineered viruses into the wild.” DARPA stated in writing that it was not. The February 2019 announcement says “DARPA is not funding the release of PREEMPT interventions into the environment”, and the programme page says the research is performed entirely in controlled laboratory facilities. The legitimate criticism is different and sharper: that page refers potential future field trials to “other government stakeholders” without naming any — a referral to an authority that does not exist.

Frontier “Transmissible and transferable are the same thing.” They are the distinction most of the policy turns on. A transferable design moves by contact and stops; a transmissible design moves by infection and continues. The people most opposed to the transmissible route — Sandbrink, Watson, Hebbeler and Esvelt in 2021 — recommended the transferable route as the alternative. Collapsing the two turns a disagreement about reversibility into a disagreement about vaccines.

Speculative “Modelling shows large reductions in disease, so the benefit is established.” The 2024 Expert Review of Vaccines paper, written largely by the technology’s proponents, states that superinfection rates are unknown and that the dynamics of the relevant vector class “remain almost completely obscure”. Model outputs under those conditions describe the consequences of assumptions — a legitimate and useful thing to do, and not evidence of effect.

Frontier “It would be contained because it would go into a population that already carries the vector.” On the proponents’ own analysis that is true and it is the problem. Their 2024 conclusion is that such a design is not expected to expand its footprint of active infection, so coverage would need repeated releases or release into a naive population, where “containment then becomes a concern”. Containment and efficacy pull against each other, and a proposal has to say which it is choosing.

Speculative “Reversion is an engineering problem that has been solved.” No — it is an open evolutionary question, and this brief deliberately does not discuss the mechanisms. At the governance level: selection in a wild population is not under the releaser’s control, the relevant rates have never been measured in a free-living population of any engineered self-spreading agent, and the one deliberately released virus with a seventy-year field record coevolved with its host and produced a naturally recombinant form in a different genus. Anyone asserting the problem is solved is asserting something no measurement supports.

Handwave “Someone will do it anyway, so we should stay ahead.” This is the weakest load-bearing claim in the debate and it appears on the permissive side of almost every discussion. It names no actor, offers no assessment of what one would need, and does not address whether a permissive posture in one jurisdiction raises or lowers the probability of a release elsewhere. It should be answered the way any other empirical claim is answered, and it has not been.