Inasmuch as this work has caught public attention, it has been through spectacular announcements about, say, artificial organisms, or concerns over the potential for synthetic biotoxins. Most of this work is neither spectacular nor reckless, however: it is about fine-tuning genetic circuits and metabolic pathways in microbes that never leave the fermenters where they are grown. It is being done in service of life-saving medicines, more sustainable materials, ecological renewal and cleaner fuels. And the field is not monolithic: plant, animal, microbial, environmental and therapeutic synthetic biology differ in their methods, their risks and their governance.
Despite this granularity, the whole field is increasingly being judged and governed with reference to a very particular case: engineered life that is meant to persist, spread or act at scale. Engineered organisms do grow and reproduce, and they can be designed to persist, though most struggle outside the laboratory. To keep an engineered organism in hand, one must work against the organism itself or its most basic tendencies: against its capacity to reproduce, to mutate, and to recover by selection any capabilities that have been deliberately removed from it. This is not a judgement on the organism; it is the field insuring itself against its own material – responsible design rendered in molecular terms. The material does not merely wear or fail, as an engineered thing might; it adapts. This is a design problem peculiar to a medium that is alive.
Years of deliberate safety design have built up a suite of measures designed to tackle this peculiarity, including kill switches, gene-flow barriers and synthetic dependencies. A key safety measure, first realised at genome scale a decade ago, is auxotrophy, which means ensuring that an engineered organism requires a specific molecule to live – a molecule that occurs nowhere in nature. The organism therefore lives at our discretion: inside the bioreactor it thrives, outside it cannot survive. Because a mutation might remove this dependency, more robust designs layer several such dependencies, each working by a different mechanism, into a fail-safe stack so that no one accident undoes them all. Even so, anxieties about the technology persist, and will probably continue to do so, because as examples like this illustrate, the opportunities and challenges of designing, building and programming life are quite distinct from those of engineering inorganic matter.
Programmable matter and its trace
The idea of life as programmable matter invites a picture of the material as specifiable: something that can be written to, made to hold a function, treated as defined by what it is engineered to do. Yet, any change made to a living system leaves a trace, and that trace may have long-lasting consequences. Changes written into the organisms’ genetic material can be inherited, and inheritance is never perfectly faithful: it introduces variations that no one specified – not beyond all control, but never fully under control either. What is being engineered is not an object, or an individual organism, but a lineage. The sequence of states through which a living system has passed partly constitutes what has become; take a different path and you would end up with a different being. This is not a limitation of present technique, to be dissolved by better tools. It is how living systems are, whether the most carefully-specified, lab-engineered minimal genome or a wild population.
Much of the biosecurity conversation turns on the idea that an intervention that has gone badly wrong could be withdrawn and the prior situation restored. Some designs are indeed intended to be reversible, and where reversal is genuinely deliverable, it deserves its place in governance. But the difficulty is that one rarely knows in advance whether this reversibility can be achieved in practice. Even if the organism can be contained or eradicated, the consequences of its release may continue to rebound – for example, if it has already tipped an ecosystem in a particular direction. So a release decision merits full scrutiny, whatever safety features have been written into the organism, and all governance decisions must be provisional, with regulations periodically reviewed, revised and, where warranted, relaxed.
What reversibility cannot underwrite, two companion capacities must cover: remediability – the capacity to mitigate harm, alongside efforts to undo previous designs; and navigability – the capacity to steer a system forward. These capacities can also be incorporated into the fail-safe stack. But neither restores the previous states of the system; both shape what comes next. Practitioners have long known this in practice long, before saying so in concept. The fail-safe stack does not restore a reactor or a soil to a prior state, or attempt to. Instead, it keeps the organism on a trajectory within a range that can still be influenced – steering coded into a genome at design time, before the organism actually exists. Biocontainment is not a safety feature bolted onto programmable matter. It is the concession, written in molecular terms, that the ability to recall the organism, or reverse its effects, cannot be assumed.
This perspective leads us to ask different questions about the effectiveness of biosecurity in synthetic biology. If containment were a safety feature, the question would be how well it works, and the answer would be a probability of escape. If containment is steering capacity rendered in a genome, the questions are multiplied and more nuanced. Over what period must that capacity persist – an experiment, a growing season, an evolutionary timescale? What selection pressures might undo it? An auxotrophic dependency undermines the fitness of an organism, so natural selection will inevitably tend to route around it. Who holds the steering capacity once the organism is deployed, and will they still exist in 10 years? Fail-safe stacks are a design-time answer to a question that goes on being asked long after design is finished, by something that is reproducing, mutating and exchanging material with whatever it meets. An engineered organism, as noted earlier, is not a device that has been placed somewhere. It is a lineage that has been started.
That a lineage cannot be secured, the way an object can, has already become evident in what the field is pursuing. Alongside containment, researchers are exploring organisms designed to exhibit goal-seeking and problem-solving behaviour, and evolutionary processes designed to remain open-ended. Neither is incompatible with containment in any narrow sense, but they assume different things about the material: whereas containment treats it as something to be kept within bounds, open-endedness and problem-solving prize its capacity to move beyond them. How these two commitments are meant to coexist is not obvious, and the difficulty is rooted in how we conceptualise the challenge.
The vocabulary of “containment” construes the challenge as the organism “escaping” from its desired state – gaining unanticipated capabilities or migrating out of a specified environment. But this framing is an interpretive choice, not something the biological reality dictates. It presupposes what we want is for engineered life to stay put – that the reactor wall marks the boundary of legitimate action. In fact, much of today’s synthetic biology does have this ambition: a cellular factory should stay in its vessel, and containment of the ordinary kinds is well matched to that ambition. The field has not been complacent about it. The “beyond containment” conversation within the biosafety community has widened the safety repertoire well past physical barriers, to kill switches, synthetic dependencies and genetic firewalls. But that repertoire still answers the question of how to keep engineered life in check. A growing share of what the field now expects of itself asks a different question, because the aim is not to keep the organism in but to put it out: synthetic biology can deliver some of its social and ecological promises only if certain releases are made responsibly rather than prevented.
When release is the point
The sharpest instance is already in development with gene drives: edits engineered to override the ordinary odds of inheritance, so that they propagate through a wild population faster than selection removes them. One potential application, for example, is to introduce a genetic modification that makes mosquitoes less capable of hosting the parasite that causes malaria, or which makes them less capable of reproducing; in both cases, the rapid spread of this edit could significantly decrease the spread of malaria. Gene drives can be modulated: self-limiting designs such as “daisy-chain drives” exhaust themselves after a few generations, and “reversal drives” are being developed which would overwrite earlier edits, though both variants remain unproven at scale and are in turn subject to considerations of mutation and selection.
With genetic edits designed to spread, new challenges emerge. A mosquito recognises no jurisdiction, so the release of gene drive mosquitoes by one authority will have consequences for many who do not have any affiliation with that body. One response is to bring together those who might be affected in a coalition of the willing, but the real challenge is the mismatch between the shape of the technology and the shape of the institutions that would govern it. Work has begun on reshaping those institutions: for example, the Convention on Biological Diversity has convened expert groups on synthetic biology and gene drives, the World Health Organization has issued guidance on testing genetically modified mosquitoes, and programmes such as Target Malaria now engage affected communities years before any release.
The mismatch persists, but it is being worked on – and the need for this work will continue as synthetic biology continues to advance. The unit of design will likely expand from molecules to cells, then to organisms, populations, and eventually entire ecosystems. We might credibly envisage corals engineered for heat tolerance and returned to the reef, microbes released to digest pollutants, engineered organisms introduced to restore biodiversity in degraded ecosystems and living systems enlisted in geoengineering the climate – a prospect that raises governance questions of its own. Somewhere along this roadmap, containment ceases to be a strategy that might fail and instead becomes a category that does not apply: there is no fail-safe stack to build into a reef.
But that is not to say that there is nothing exerting control over the released organism. When an engineered microbe enters a soil, it is recognised by immune systems, infected by phages, eaten by predators, broken down by decomposers and outcompeted (or not) by whatever else lives there. It becomes entangled in an ecological web, subject to innumerable relations that act on it. Its designers may not have anticipated the niche it comes to occupy, and it may do considerable harm, but this will be constrained by an ecological firewall: a steering capacity of a kind supplied by the living world upon release, rather than written into it at design time. Release, therefore, does not abolish navigability. It transfers it – to dynamics no one steers, and for which no one answers.
Which is what makes the ecosystem-scale ambition different in kind rather than degree. When the ecosystem itself becomes the object of design, what had been doing the steering all along is the thing being redesigned. The question stops being whether the organism stays where we put it, and becomes whether anything in the environment is still positioned to answer back. Again, there is a conceptual issue here: the field’s own talk of “stewardship” or “partnership with the planet” sits uneasily beside an activity that specifies in advance what an ecosystem is for. Reaching for both partnership and ecosystem design at once claims the moral standing of the first with the operational authority of the second. This is not a confusion so much as a sign that a field has outgrown the available language. Yet the two are not irreconcilable: stewardship traditions – not least Indigenous, local and traditional ecological knowledge – have long tended living systems without claiming design authority over them, which suggests partnership could discipline design rather than merely decorate it.
What makes this pressing, rather than merely conceptually interesting, is timing. An oft-cited precedent is the 1975 Asilomar Conference on Recombinant DNA, which brought together scientists, lawyers and medical practitioners to draw up voluntary guidelines for research in the emerging field of biotechnology. While its success is contested – commercial development of recombinant DNA accelerated almost immediately afterwards – Asilomar illustrates how a field can effectively deliberate about its own constraints when it is small enough for a group of practitioners to see the whole picture. Once a domain has mature techniques, trained communities, tacit know-how and commercial incentives, the same conversation is harder to convene and weaker in what it can ask – not because anyone acts badly, but because a research community is a set of careers, investments and skills that cannot simply be stood down. Mature fields are not ungovernable – clinical medicine is among the most tightly governed activities there is – but they are governed in terms of what they have already become, not what they might one day become. What closes as a field matures is not regulation but the chance to shape what it will be, and it closes fastest when capability becomes commercially or strategically valuable.
How to approach anticipatory governance can be framed as emerging from a series of attitudes, each taking over when the previous one has reached its limits.
A starting point is prohibition – not opposition to the field in general, but specific judgements about modifications or interventions that should not be made at all and thresholds that should not be crossed. Gene-editing human embryos for enhancement purposes, for example, is currently in this category. It is comparatively easy to hold the line on such prohibitions when reliable techniques do not exist; once they do, incentives harden and prohibitions become harder to maintain. This has been a recurring instrument of life-science governance, and has held without any global regulatory authority. The experiments that the Asilomar participants agreed to forgo, the near-universal concerns on heritable genome editing, successive moratoria on various technologies – all have been enforced by funders, journals and professional norms.
But prohibitions can only address developments that are already clearly anticipated, and synthetic biology is a fast-evolving field. The attitude that survives this objection is steering – a readiness to grant that control of the old kind has already been outrun, and accept that task now is to shape trajectories as they unfold, by improving monitoring, developing more effective mitigant strategies and building capacity to change course. But steering assumes agreement on who steers, on what basis and towards what; and such agreement is hard to secure when there are many actors pursuing different ends for different reasons.
What both prohibition and steering reach for, and neither secures, is a relationship – a standing arrangement between the parties involved, of the kind that lets a threshold be held or a direction chosen together, which exists before the moment it is needed. Prohibitions hold only because of who stands behind them, and where a community is dense enough to notice and sanction defection. But as the field grows more distributed, it becomes less obvious who stands behind them, or what the cost of breaking the prohibitions would be. Synthetic biology currently stands on the cusp of this transition, making it a matter of some urgency that these relationships be built now.
The logic of governance reverses at this point. These attitudes treat the engineering of biology as the thing to be governed, and a relationship between stakeholders as the means of governing it. But the need of a field for sustained cooperation across borders, sectors and disciplines is not merely a matter of governance; it is the basis of the mutual dependence that binds differentiated societies together more reliably than shared values ever have. Whether that is grounds for hope, or just a way of making a virtue of a necessity, cannot be settled here, but it is a possibility that is hard to see, much less realise, when governance is fixated on control. This moves the question from what engineering biology will make of us and our world to how we ought to act in it – from what we are to how we conduct ourselves.
No single vantage point
The more synthetic biology advances, then, the more pressing the need for governance based on relationship-building, rather than attempting to pre-determine outcomes at the design stage. This need is being further exacerbated by the advent of artificial intelligence, which is not just facilitating but driving synthetic biology. AI can be used both to analyse genomic data at scale, and also to design and direct modifications to it. An AI-designed organism can, of course, be tested against the usual biological criteria – toxicity, fitness, off-target effects – and will be. But it still creates new challenges for governance.
The design rationale, including the logic of any safeguards, will be undocumented, weakening accountability when something needs explaining. Generative design expands the space of candidate sequences beyond known threat signatures, making it harder to screen for problematic modifications. And while direct testing might catch most threats, what it cannot anticipate is judged against standards that are themselves provisional – the benchmarks for what counts as a safe or correct design shift as the science advances. Confidence will be set by the weaker of the two: the AI model that produced the design, and our evolving grasp of what to test it against. Biosafety and AI safety are therefore becoming inseparable at the design layer, dual use included: the architecture that designs a pollution-degrading enzyme could also be pointed at designing a pathogen. Dual use here is not merely a misuse scenario appended to the technology, but instead the same technology, seen from another angle.
Another challenge stemming from the relational perspective concerns distribution. Generic platforms, cheap synthesis and cloud laboratories are being pursued precisely so that biological design capabilities do not end up confined to a few well-resourced institutions. Again, this has both positive and negative aspects. Equitable access means giving access to ignorant and malicious actors as well as to those who would otherwise be under-privileged. That comes with risks. But it is not obvious that concentrating capability in the hands of whoever already owns the apparatus is necessarily safer. And some of the infrastructure needed to manage such risks already exists in early stage: the synthesis industry’s screening consortium, the safety norms built into competitions such as iGEM, the shared codes of conduct of community laboratories. Relationship-based governance is already operating at the distributed end of the field. But it does mean the system acquires no vantage point. It runs through laboratories, companies, funders, databases and automated platforms that no national regulator sees in their entirety, while those affected include patients, farmers, workers, future generations and nonhuman life, none of whom are party to the design decisions.
When no one holds the whole picture, the reflex is to ask whether the actors can be trusted. That is not wrong so much as misdirected. Trustworthiness is a property of parties; what is needed is a property of the relationships between them, and the second is not produced by securing the first. This matters because disagreement here rarely comes from bad faith. It comes from actors pursuing what each takes to be desirable ends and differing over what those ends are – a plurality that will not dissolve under better evidence. The strain shows where the relationships are: states questioning each other’s intentions; academic and commercial incentives generating mutual suspicion between people ostensibly working on the same problem; disciplines contesting who has the standing to say what a good outcome would be. None of these is a shortage of goodwill.
This is why standing institutional reflection matters. The field has produced it before, when at its best: the International Summits on Human Genome Editing, the World Health Organization's governance framework and global registry, the Global Alliance for Genomics and Health, and the expert groups convened under the Convention on Biological Diversity are all – whatever their imperfections – places where questions are held rather than cleared, and they are more inclusive than anything the field has managed before. The precedents that worked in this field worked that way. The containment practices of 1975 were adoptable because a community deliberated before expanding. The decision to treat genomic sequence data as a shared resource was a decision about what relationship the participants would stand in, taken while it could still be taken. Neither the containment practices nor the resource-sharing were purely risk calculations; each was also an act of deciding, together, what kind of field to build.
This changes how governance is built. If governance capacity is relational capacity, it will have to be built like any other relationship: slowly, before it is needed, by parties who expect to deal with each other again. It cannot be abruptly convened after an incident; it cannot be delegated to whatever regulator is deemed to have jurisdiction, because none has enough; and it cannot be assembled on the basis of technical expertise alone, because the disagreements it must hold are not purely or even primarily technical. All this gives science diplomacy a more demanding role than the one it is usually assigned. Not diplomacy in support of science – facilitating collaboration, easing the movement of people, knowledge and materials, with science as the end it serves – but diplomacy whose object is securing the relationships themselves, between countries, disciplines, sectors and publics, with the work of science as one of the places those relationships are made, and treated as the substance of governance rather than its precondition.
What this asks is concrete, and different for each stakeholder community. Funders can treat relationship infrastructure – standing forums, joint monitoring, shared registries – as seriously as platforms and pipelines. Diplomats can treat convening capacity as governance capacity, built before incidents rather than after them. Regulators can plan for remediation and course-correction, not only for approval or refusal. Researchers can treat dual-use awareness and engagement with affected communities as design-stage activities rather than after-thoughts. And all of them can act while the field is still young enough for these relationships to shape what it becomes.
What recurs across reversal, containment, boundary-setting and steering is the same discovery in a different place: that the control being reached for is not quite available, and that what remains when it runs out is a relationship – to the living systems being altered, and to one another. That relationship is not something to be secured before the work proceeds. It is the medium the work is already being done in.
