Biological Robots

Frontier The prize is a repair agent a patient’s own body could grow without a single genetic edit: a motile construct that assembles itself from unmodified cells and then does a job the source genome never specified. The one measured therapeutic result is a scratch assay — concentrated assemblies of Anthrobots, built from adult human tracheal epithelial cells, sat on a wound cut into a cultured neuron monolayer and a bridge of neurons as thick as the surrounding healthy tissue grew back underneath them, while the same wound left uncovered showed none.

Established Stated in the units that are actually measured, the objects are small, short-lived and built from nothing exotic. Anthrobots are 30 to 500 micrometres across, self-assemble from cells that spontaneously form organoids, move for 45 to 60 days on their own cilia and then biodegrade; xenobots, their frog-cell predecessors, close a laceration spanning about half their diameter within 5 minutes in every one of 15 injured individuals, with no deaths. Frontier Cells taken to day 10 as Anthrobots even read about 4.6 years younger on a DNA-methylation clock than the progenitor cells they were made from — roughly a 25 per cent drop in epigenetic age, from a donor of 21.

Speculative Who gets this, and when, is where the honesty has to start. The constructs are built from a patient’s own tissue, so the autologous, no-transgene route sidesteps both rejection and the containment problem that stops released engineered organisms — but every therapeutic claim here is an in-vitro observation, no construct has been tested in a living animal, and the field’s own leaders call the applications wholly speculative.

Handwave The condition on all of it is a mechanism nobody has isolated. If the living, moving construct adds something a smear of dead cells or secreted factors would not — which no blinded, independently replicated experiment has yet shown, and section 7 says exactly what would — then this is a genuinely new kind of medicine. If it does not, it is an elaborate way to deliver a conditioned-medium effect, and section 5 is where that doubt lives.

The term covers a specific and narrow thing: multicellular constructs that self-assemble from unmodified animal cells and then move and act, without a genome edited for the purpose and without a synthetic scaffold. Two systems define the category. Xenobots are built from the cells of Xenopus laevis frog embryos; Anthrobots from adult human airway cells. Neither carries an engineered gene. What makes them interesting is that they do things — travel, aggregate loose material, close their own wounds, apparently encourage tissue to regrow — that no frog and no human airway ever does, using exactly the genome a frog or a human airway already has.

This is why the subject is separate from its neighbours on the map. Designer Organisms asks whether an organism can be built to a genetic specification; here nothing is edited, and the specification, such as it is, operates on shape and culture conditions. Synthetic Biology owns standardised genetic parts. And the wider field of biohybrid robotics — muscle cells on a polymer skeleton, cardiac tissue driving a synthetic swimmer — is a different thing again: it bolts living tissue onto engineered materials, where a biological robot in this brief’s sense is cells all the way through. A 2025 review counted that biohybrid field in the thousands of papers and explicitly set the self-assembled, cell-only constructs to one side; this brief is about the set it excluded.

The framing claim, made by the constructs’ own inventors, is that the genome is better understood as hardware that can run more than one program. The cells bring competencies — motility, signalling, self-repair — that were selected for one body plan and turn out to be reusable in another. Whether that reframing is deep or merely a vivid way to describe developmental plasticity is one of the questions section 4 has to weigh.

Established The first xenobots were designed in a computer and built by hand. Kriegman, Blackiston, Levin and Bongard reported in PNAS in 2020 that an evolutionary algorithm searched in simulation for body shapes that would move, and the winning designs were then assembled by hand from Xenopus laevis skin and cardiac cells into sub-millimetre constructs that crawled using beating cardiac tissue. Frontier The claim that made news — that a computer had “designed” an organism — is weaker than it sounds: the algorithm chose an outline, and biology supplied everything about how the cells actually behaved once placed together.

Established The second generation dropped the hand-sculpting and let the cells build themselves, and this is where the measurements get solid. Blackiston, Lederer, Kriegman, Garnier, Bongard and Levin reported in Science Robotics in 2021 that explants of animal-cap tissue from stage-9 Xenopus embryos — spheres of 3,026 ± 180 cells — self-organised over about four days into ciliated constructs 487 to 602 micrometres across that swam by ciliary beating at speeds >100 micrometres per second. Established They confirmed no neurons were present (n = 24), so whatever the constructs do is done without a nervous system.

Established What those self-assembled xenobots do is a real behavioural repertoire, quantified. Cross-entropy clustering sorted their movement into four discrete states; they traversed open fields, 1-centimetre mazes, 2-millimetre passages and narrow capillaries, with 42 per cent (n = 12) passing through a capillary tube end to end. Released among 5-to-10-micrometre iron-oxide particles, multiple xenobots spontaneously pushed the debris into piles over about 12 hours — a collective behaviour a physics model reproduced with no sensing required. Frontier The particle-aggregation result is the one most often narrated as intelligence; the modelling that reproduces it without sensory feedback is the reason to keep the flag at frontier rather than higher.

Established Their self-repair is the least ambiguous result in the whole subject. After lacerations spanning roughly half the construct’s diameter, all 15 injured xenobots closed the wound within the first 5 minutes of observation, with no mortality and statistically significant shape recovery (P < 0.0001). Lifespan was about 9 to 10 days on the yolk the embryonic cells carry with them, extendable beyond 90 days with added media, and swimming speed did not decline with age (P = 0.29).

Frontier The result that generated the most confusion is kinematic self-replication. Kriegman, Blackiston, Levin and Bongard reported in PNAS in 2021 that a parent sphere of about 3,000 cells, moving through a dish of loose dissociated stem cells, pushes those cells into piles that mature into new xenobots. Established Unaided spheres manage one or two such generations and then the process dies out; an evolutionary search on a supercomputer found that a C-shape, like a flattened Pac-Man, piles cells far more efficiently and sustained four to five successive generations. Frontier The mechanism is entirely mechanical and depends on an external supply of loose cells; it copies a shape, not a genome, and transmits no heritable information. Calling it self-replication is defensible only with that qualifier attached, and section 15 is where the unqualified version is answered.

Frontier Anthrobots moved the result from frog embryos to adult human cells, which is the anchor of this brief. Gumuskaya, Blackiston, Levin and colleagues reported in Advanced Science in 2023 that adult human tracheal (airway) epithelial cells, given culture conditions that turn their cilia to face outward, spontaneously form motile multicellular constructs with no genetic modification of any kind — no edits, additions or deletions. Established The constructs range from 30 to 500 micrometres, survive 45 to 60 days before biodegrading, and fall into morphological types whose shape predicts their movement: fully ciliated spheres wiggle in place, while irregular, patchily ciliated forms travel in straight or curved lines. Separate Anthrobots also fuse spontaneously into larger assemblies the authors call superbots.

Frontier The therapeutic observation is a single in-vitro assay, and its exact shape matters for how much weight it can carry. A dense assembly of Anthrobots was placed across a scratch scored into a monolayer of cultured human neurons; neurons regrew under the assembly to form a bridge as thick as the surrounding healthy layer, and an identical wound with no Anthrobots showed no regrowth. Frontier One control was run and it is an important one: an inert polysaccharide gel produced no bridging, so the effect is not the construct merely acting as a physical scaffold. Handwave What was not run is any control isolating secreted factors — conditioned medium, cell-free debris or a dead construct — and the lead author states plainly that the mechanism is unknown. The result is real and surprising; the gap between it and a therapy is the subject of section 7.

Frontier A 2025 follow-up added a measurement nobody predicted: the constructs appear to get younger. Gumuskaya and colleagues, in Advanced Science, put DNA-methylation clocks on Anthrobots from a 21-year-old donor: progenitor cells read a tissue age of 23.7 years, day-10 Anthrobots 19.1 years (about 4.6 years lower, P = 0.0015) and day-25 constructs 20.0 years (about 3.7 years lower, P = 0.0087) — roughly a 25 per cent reduction in epigenetic age. Speculative It happened without any up-regulation of the Yamanaka reprogramming factors, so the usual explanation does not apply, the sample sizes are small (n = 3 at day 25), and the authors do not claim a mechanism. It is a striking correlation in search of a cause, not an anti-ageing result.

Frontier The newest work, in 2026, put neurons into the constructs. Fotowat, Levin and colleagues reported in Advanced Science that neuronal precursor cells implanted into biobots during formation differentiated into mature neurons with axons and dendrites, self-organised into networks, and made the constructs more elongated and behaviourally more active; a seizure-inducing drug, pentylenetetrazole, changed their movement, which is evidence the neural activity is doing something. Speculative Genes for visual-system development were up-regulated and the authors raise, explicitly as a hypothesis needing protein-level confirmation, that “some kind of visual system may be developing.” Frontier This is the direction the field is moving — toward constructs that sense and act — and it is at the stage of first demonstrations, not capability.

Established The honest summary of the position is that the self-assembly and the motility are solid, and everything downstream of them is younger than the headlines. Constructs self-build from unmodified cells: measured and replicated across two species and multiple labs’ methods. They move and self-repair: measured. They do something to a wounded neuron layer in a dish: observed once, under-controlled, mechanism unknown. They are useful medicine: not shown in any living animal.

Handwave The load-bearing open question is whether a living biological robot does anything its own dead cells or secretions would not. The neuron-bridging result has exactly one published control — the inert gel — and it rules out only the crudest alternative. Until a cell-free-debris arm, a dead-construct arm and conditioned medium have been tested against the living construct, blinded and by an independent group, the phenomenon is compatible with a plain paracrine effect delivered by a novelly-shaped clump of cells.

Frontier The conceptual claim — that these constructs show behaviour the genome does not specify — is real but easy to overstate. Every capability on display is one the cells already had: airway cilia beat to clear mucus, embryonic cells migrate and heal. Frontier What is genuinely unspecified is the whole-construct behaviour that emerges when those competencies are reassembled in a body plan evolution never built — and whether that is a deep fact about cellular agency or a restatement of developmental plasticity is contested among the very authors who raise it.

Frontier Whether shape can be used as a control interface is the field’s most practical open question. In xenobots and Anthrobots alike, morphology predicts movement, and in the self-replication work an AI-chosen shape multiplied the number of replication rounds. Speculative If form reliably programs function, then culture conditions and geometry become the “code” for these objects; whether that mapping is precise enough to engineer against, rather than merely observe, is unshown.

Frontier Kinematic self-replication raises a live question that the semantics usually bury: how far can the copied shape carry function across generations? Four to five generations were sustained with an optimised parent and a constant external cell supply. Speculative Whether any useful behaviour — not just gross shape — can be made to persist or improve across those generations, and whether the process ever proceeds without a human refilling the dish of loose cells, are both unanswered and would change how seriously the “replication” framing should be taken.

Speculative The neurobot result opens the deepest question and the field is not close to answering it. A construct with a self-organised nervous system that responds to a neuroactive drug is a platform on which the words sensing, learning and even rudimentary experience stop being obviously inapplicable. Handwave Whether anything worth calling sensing is present, rather than gene-expression signatures and altered motility, is undemonstrated, and the visual-system suggestion is explicitly a hypothesis.

Frontier Reproducibility is a frontier question here in a way it is not for most engineered systems. As the biohybrid-robotics ethics literature puts it, two such constructs made by identical processes are not guaranteed to be even nominally identical — a consequence of building with cells that have their own dynamics. Frontier Whether construct-to-construct variability can be brought under the control that any medical or engineering use would require is unsettled and rarely quantified.

Handwave The binding bottleneck is mechanistic: nobody can say what the therapeutic effect is. A result whose mechanism is unknown cannot be dosed, optimised or predicted, and the one in-vitro effect the field rests its medical hopes on — neuron bridging — has never been separated from the possibility that secreted factors do the work. This is not a scale-up problem waiting behind a solved science; it is the science.

Frontier The second bottleneck is variability, and it is intrinsic rather than incidental. Constructs self-assemble, so their exact size, shape and ciliation are set by the cells, not the maker; morphology determines behaviour, so that variability propagates straight into what the construct does. A manufacturing process that cannot specify its product’s shape cannot specify its function.

Frontier The third is the leap from dish to body, which nothing in the record has attempted. Every therapeutic observation is in vitro. A construct built for a still culture dish faces, in living tissue, an immune context, mechanical forces, a three-dimensional matrix and a nutrient supply none of the experiments included — and the constructs biodegrade in 45 to 60 days, which is either a safety feature or a durability ceiling depending on the application.

Frontier The fourth is steering. Xenobots navigate mazes but are not directed; they go where their shape and the fluid take them. Speculative A repair agent that cannot be guided to the site of injury, held there, and switched off is a demonstration rather than a tool, and no reliable control channel — chemical, bioelectric or geometric — has been shown.

Frontier The fifth is sourcing. Anthrobots’ appeal is that they can be built from a patient’s own airway cells, but donor-to-donor variability, the yield of usable progenitor cells, and batch consistency at anything above bench scale are unaddressed. Established The frog constructs sidestep this by using embryonic cells with their own yolk supply, which is exactly why the frog results are cleaner and the human ones more consequential.

Frontier This subject waits above all on an understanding of morphogenesis — how cells decide what to build — which is a result Morphogenetic Pattern Formation owns and has not delivered. Biological robots are an application of morphological plasticity: they work because cells will assemble into forms other than the one they came from. Predicting or specifying those forms is the missing capability, and it is recorded there rather than duplicated here.

Frontier It waits on control of form, which the bioelectric programme is trying to supply. The same laboratory tradition argues that bioelectric signalling is a control layer for anatomy; Bioelectric Medicine is where that claim is assessed. If form can be steered bioelectrically, it becomes the natural interface for building and directing these constructs; if it cannot, geometry and culture conditions are the only levers left.

Handwave The sharpest dependency is a mechanistic assay that does not yet exist, recorded as a typed non-brief constraint in section 14: without a way to measure what the living construct adds beyond its secretions, none of the medical claims can be either built on or dismissed. Frontier This is a scientific result nobody is currently producing, not an institution that could choose to create it.

Speculative The neurobot direction adds a dependency on the science of self-organising neural tissue, which it shares with Organoid Intelligence: whether an implanted neural network in a construct can be read, stimulated and understood at all draws on the same unsolved problems of interfacing with small living neural systems.

Frontier The decisive experiment is a blinded, independently replicated wound-repair assay with cell-free-debris and dead-construct controls, quantifying what the living construct adds beyond secreted factors. It is the cheapest experiment in this brief, needs no new hardware, and could be completed within a year or two using assays that already exist — and no group has published it. Handwave Everything the medical case rests on turns on its outcome: if the living, motile construct outperforms a dead one and outperforms its own conditioned medium, biological robots are a new modality; if it does not, the neuron-bridging result is a paracrine effect in unusual packaging.

Frontier Second: put a construct into a living animal with a controlled comparator. The entire therapeutic literature is in vitro. A wound or nerve-injury model comparing living constructs, dead constructs and vehicle, with a pre-registered endpoint, would establish whether anything survives contact with an immune system and a three-dimensional tissue — and would be the first datum on the dish-to-body question that section 5 names as binding.

Frontier Third: an independent replication of the Anthrobot result in a second laboratory, from a different donor. The anchor results come from one research group. Given that the constructs are intrinsically variable and the effect’s mechanism is unknown, independent reconstruction — does a different lab, different donor, different technician get motile constructs and a bridging effect at all — is the standard the field has not yet had to meet.

Speculative Fourth: test whether morphology genuinely programs function, prospectively. The claim that shape predicts behaviour is drawn from observed constructs sorted after the fact. Designing a target morphology, building constructs to it, and predicting their movement in advance would convert a correlation into an engineering rule — and the self-replication work, where an AI-chosen shape raised the generation count, is the closest anyone has come.

Speculative Fifth: establish whether the epigenetic-age reduction is real and causal. The 25 per cent drop in methylation age was measured in three day-25 samples with no mechanism and no Yamanaka involvement. A properly powered replication across donors and clocks, with a functional readout of whether “younger” cells behave younger, would settle whether this is a rejuvenation phenomenon or a methylation artefact of the self-assembly process.

Speculative Sixth, for neurobots: show that the nervous system does perceptual or learning work. A drug changing movement is evidence of neural activity, not of sensing. A construct that alters behaviour in response to a controlled stimulus — light, if the visual-system genes mean anything — under blinding, would be the first evidence that these objects sense rather than merely twitch.

Established The core engineering trick is to do almost no engineering: supply culture conditions and let the cells build the object. Anthrobots are made by growing airway cells into organoids and adjusting conditions so cilia face outward; xenobots by letting embryonic explants round up. No scaffold, no printing, no genetic construct — which is the discipline’s genuine novelty and the reason it is not biohybrid robotics.

Frontier The absence of genetic modification is an engineering advantage that the field under-advertises. A construct from a patient’s own unmodified cells has no transgene to regulate, no rejection to suppress, and — because it biodegrades in weeks — no persistence to contain. Frontier That erases most of the regulatory and ecological burden that stops released engineered organisms, and it is the strongest structural reason to take the medical direction seriously despite the thin evidence.

Frontier The unsolved engineering is control: building to a specified shape, steering to a target, and switching off. Because behaviour follows morphology and morphology is set by self-assembly, a real manufacturing process needs to constrain shape reproducibly — and none is demonstrated. Speculative Directed control, whether by chemical gradient, bioelectric cue drawn from Bioelectric Medicine, or geometric templating, is at the concept stage.

Speculative Adding sensing and actuation is the engineering frontier, and neurobots are its first move. Implanting neural tissue to give a construct an internal controller is a route to autonomy that overlaps the goals of Embodied AI and Robot Foundation Models from the biological side, and it inherits every difficulty of interfacing with living neurons. Frontier What has been built is a construct whose implanted neurons fire and change its motion; what has not is any construct that uses that firing to do a task.

Established Scale and shelf life are real constraints stated in the numbers. Constructs are 30 to 500 micrometres and last 45 to 60 days (frog constructs 9 to 10 days on yolk, longer with media). Frontier Any application needs many constructs, consistent between batches, delivered within their lifespan — a manufacturing and logistics problem that no one has had to solve because no application has reached that stage.

The closest neighbour is the one this brief is most often confused with. Biohybrid robotics — living muscle or cardiac tissue mounted on synthetic structures to make walkers, swimmers, grippers and pumps — is a large and quantitatively advanced field, with cardiac-driven walkers at 8 millimetres per minute and biohybrid grippers exerting on the order of 8 millinewtons; it is reviewed as a separate discipline and it explicitly excludes the cell-only, self-assembled constructs this brief covers. The distinction is not pedantic: biohybrid devices are engineered and reproducible in the ordinary way, and biological robots are neither.

Morphogenetic Pattern Formation owns the plasticity these constructs exploit; Bioelectric Medicine the control-of-form claim; Basal Cognition the argument that cells and tissues have competencies worth calling cognitive, which is the framing under which “robot” is applied here. Organoid Intelligence is the neurobot direction’s sibling, and Embodied AI and Robot Foundation Models the synthetic-side counterpart to giving a construct a controller.

On the medical side, Limb Regeneration and Whole Organ Regeneration are the regenerative-medicine goals biological robots are pitched at, Cellular Rejuvenation inherits the epigenetic-age result directly, and Lab-Grown Organs and Nanomedicine are the alternative routes to targeted repair against which this one has to justify itself. Designer Organisms is the genetic-specification neighbour, and this brief is deliberately the non-genetic case.

Established There is no regulatory pathway for a self-assembled living construct, because it fits no existing category. It is not a drug, not a device, and not conventionally a cell therapy; it is a motile multicellular object built from a patient’s cells with no genetic change. No regulator has classified one, and nothing in the record has been submitted for approval — the work is entirely pre-clinical.

Frontier The governance conversation is running far behind the science, and this is measured, not asserted. A 2024 analysis in PNAS found roughly 1,648 biohybrid-robotics publications by May 2023 and only 5 that substantively addressed ethics. Established The same paper proposes six governance necessities — risk assessment across domains, explicit attention to social justice, an appeal to the common good including ecosystems, interdisciplinary work on moral status, stakeholder participation and public literacy — none of which has institutional form yet.

Speculative The proposals that exist treat ethics as coordination rather than constraint. A 2026 governance framework argues that the problem is not that researchers reject ethical principles but that they cannot implement them without common knowledge and mutual assurance, and proposes public commitments, ethical observatories and interoperable governance infrastructure. Frontier These are proposals, not institutions, and they are being written for a technology that has not yet produced a product to govern.

Frontier The institutional finding that generalises is that the absence of a category cuts both ways. It removes the transgene and containment burdens that stall released engineered organisms, and it also removes the settled rules, endpoints and liabilities that a therapy needs to reach patients. A construct that is neither drug nor device nor organism has an open road and no map.

Frontier The moral-status question is unavoidable once neurons enter the construct, and the honest position is that no one knows where the line is. A ciliated frog-cell sphere is uncontroversial; a construct with a self-organised nervous system that responds to neuroactive drugs is not, and the field has moved toward the second in a single research programme. Handwave Whether any current construct has interests that could be harmed is undemonstrated and, on present evidence, unlikely — but the trajectory is toward exactly the properties that would make the question real.

Frontier The human-cell provenance carries its own consent question. Anthrobots are built from donated human tissue, and a construct made from a person’s cells that then moves, self-repairs and reads younger than its source is not obviously covered by a standard tissue-donation consent form. Speculative What a donor is consenting to when their cells become a motile construct is unsettled and largely undiscussed.

Speculative Self-replication is the fear that travels furthest and the one the evidence most deflates. Kinematic replication needs a human-refilled supply of loose cells, copies a shape not a genome, and peters out in a handful of generations; it is not a self-sustaining process and carries no evolutionary open-endedness. Frontier The legitimate version of the concern is precautionary and forward-looking, not a description of anything that has happened.

Speculative The dual-use and framing risks are real but should be stated at their true size. The dual-use hazard is generic to cell and tissue technology rather than specific to these constructs, and the more immediate ethical risk is misdescription: calling an under-controlled dish result a healing robot invites both overinvestment and overreaction. Established Independent researchers have already objected to the word “robot” itself, arguing the coordinated cilia are doing their ordinary job; that dispute is partly semantic and partly a real caution against reading agency into physics.

Frontier The durable idea in this subject is that the genome may be less like a blueprint and more like reconfigurable hardware, and biological robots are the cleanest demonstration of it. Unmodified cells, given a new arrangement, produce a functioning entity evolution never made. Speculative If that generalises, the long-run consequence is a form of engineering that programs living matter by shaping context rather than editing code — a different discipline from the genetic one Designer Organisms assesses, and one whose reach is entirely unknown because it rests on a plasticity nobody can yet predict.

Speculative The plausible near-term civilisational effect is medical and modest, not systemic. If the mechanism question resolves favourably, autologous, biodegradable repair constructs would be a useful addition to regenerative medicine — targeted, transient, rejection-free — competing with cell therapies and Nanomedicine rather than replacing a civilisational function. Handwave The image of “living machines that will do X” at scale has no evidential basis in the record and belongs to marketing, not measurement.

Frontier The genuinely novel long-run issue is conceptual: the category boundaries this technology crosses. An object that is alive, built, motile, self-repairing, made of a person’s cells and neither organism nor machine strains the legal, ethical and linguistic categories a society uses to decide how to treat things. Speculative That is a slow problem, and it arrives whether or not the medicine works.

These horizons track what has been demonstrated versus what would have to be shown, and the binding variable is the mechanism question rather than any engineering milestone:

• 10 yr: Frontier The decisive controlled and independently replicated wound-repair assay is run and either supports or dissolves the therapeutic claim; first in-vivo tests with proper comparators are reported. Speculative Neurobot work establishes whether an implanted network does perceptual or task work or merely fires.
• 25 yr: Speculative If the mechanism resolves favourably, autologous biodegradable repair constructs reach early clinical testing for a narrow indication; if it does not, the field remains a developmental-biology tool for probing morphological plasticity, which is a real and sufficient outcome. Frontier Reproducible manufacture to a specified shape is either solved or shown to be the limiting difficulty.
• 50 yr: Speculative Constructs with engineered sensing and steering, if the control problem yields, perform directed tasks inside the body or the environment. Handwave Programming living matter by context rather than genome becomes a named engineering discipline only if the plasticity it depends on becomes predictable, which nothing in the current record forecasts.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The one defensible statement is that the category questions — what a built, living, self-repairing artefact made of human cells is, morally and legally — outlast any particular application and will be settled by institutions, not laboratories.

• Depends on Morphogenetic Pattern Formation, which owns the developmental plasticity that lets unmodified cells assemble into forms their body never built, and Bioelectric Medicine, which owns the claim that form can be steered by bioelectric signalling. Both are results this subject consumes rather than produces; nothing else on the map blocks it.
• Requires (not on this map) Four constraints no brief here produces. The first is the decisive scientific result of section 7: a blinded, independently replicated account of what the living, motile construct adds beyond secreted factors, without which nothing medical can be built or dismissed. The second is a regulatory category, because a self-assembled living construct is neither drug, device nor conventional cell therapy and no regulator has classified one. The third is a supply chain: reproducible sourcing of patient-derived progenitor cells at more than bench scale, which donor variability and yield currently prevent. The fourth is industrial capacity to manufacture motile constructs with batch-to-batch consistency, which is hard precisely because the constructs set their own shape and their shape sets their function.
• Enables Autologous, biodegradable repair agents for regenerative medicine, in-vitro platforms for studying morphogenesis and self-organising neural tissue, and, speculatively, sensing-and-acting constructs shared with Organoid Intelligence. No typed enabling edge is claimed, because every downstream use waits on the mechanistic result above.
• Adjacent Designer Organisms, Morphogenetic Pattern Formation, Bioelectric Medicine, Basal Cognition, Organoid Intelligence, Embodied AI and Robot Foundation Models, Limb Regeneration, Whole Organ Regeneration, Cellular Rejuvenation and Lab-Grown Organs; and outside the map, biohybrid robotics, developmental biology and tissue-donation ethics.

Handwave “Xenobots reproduce.” They perform kinematic self-replication: a parent construct pushes externally supplied loose cells into piles that become new constructs. Established It requires a human-refilled supply of dissociated cells, copies a shape rather than a genome, transmits no heritable information, and dies out after one or two generations unless an AI-chosen shape pushes it to four or five. It is real and it is not reproduction in the biological sense, and the coverage that dropped the qualifier is why the claim circulates.

Frontier “Anthrobots heal wounds / are a therapy.” The evidence is one in-vitro scratch assay in a neuron dish, with a single control and an unknown mechanism, never tested in a living animal. Handwave The self-assembly is genuinely surprising; the therapeutic claim is a dish, and the field’s own leaders call clinical applications wholly speculative.

Frontier “An AI designed these organisms.” An evolutionary algorithm chose outlines that move or replicate better; the cells supplied every actual behaviour, and the result was a shape, not a program. Established Calling the constructs AI-designed overstates the algorithm’s role and understates the biology’s.

Frontier “Biological robots are a new life form.” They are unmodified frog or human cells in an unusual arrangement, running the genome those cells already had. Frontier What is new is the body plan and the behaviour that emerges from it, not the genetics or the life; “new kind of organism” is a stronger claim than the evidence, and the inventors themselves frame it as reusing existing competencies.

Speculative “Anthrobots reverse ageing.” Cells taken to day 10 read about 4.6 years younger on a methylation clock than their source, roughly a 25 per cent drop, in a handful of samples, with no mechanism and no Yamanaka-factor involvement. Frontier It is a correlation worth chasing, not a demonstrated rejuvenation, and section 7 says what would settle it.

Established “Self-healing means they regenerate like an organism.” Xenobots close a laceration within 5 minutes because their cells retain embryonic wound-closure behaviour, not because the construct regenerates a lost part. Frontier It is a real and fast result and a narrower one than “regeneration” implies.

Frontier “This is the same as biohybrid robots.” Biohybrid robots mount living tissue on synthetic scaffolds and are engineered and reproducible in the usual way; biological robots are cells all the way through, self-assembled and intrinsically variable. Established The 2025 review that surveys the biohybrid field sets these constructs explicitly outside its scope, and conflating the two borrows a reproducibility the cell-only constructs do not have.

Handwave “Living machines will soon do useful work in the body or the environment.” No construct has been steered to a target, held there, switched off, or shown to function in a living body. The gap between the demonstrations and that sentence is the whole of sections 5 and 7, and stating it as near-term is the central overclaim this brief exists to flag.