Xenobots Are Called Living Robots.
They Are Just Living.

Section One

The Version Everyone Knows

Here is the standard account, stated as fairly as possible: in January 2020, a team from Tufts University, the University of Vermont, and the Wyss Institute at Harvard announced that they had built the world's first programmable living organism. They called it a xenobot, after Xenopus laevis, the African clawed frog whose embryonic cells they used. The researchers harvested stem cells from frog embryos, then arranged them according to body plans generated by an evolutionary algorithm running on a supercomputer. The result: small (under one millimeter) multicellular blobs that could move through water, push objects, and repair themselves when cut.

The framing is not dishonest. These things were designed by a computer, not by evolution. They weren't found in nature. They move in ways their constituent cells wouldn't move on their own. "These are novel living machines," said Joshua Bongard, the computer scientist and robotics expert at the University of Vermont who co-led the research. "They're neither a traditional robot nor a known species of animal." That's a reasonable sentence. Most of the headlines stopped there and called it a day.

By 2021, the team had found something more startling: the xenobots could self-replicate. The Pac-Man-shaped parent organisms moved through their environment, gathered loose stem cells into their “mouths,” and compacted them into offspring that looked just like themselves. The AI had designed the body shape, a semi-toroidal form, specifically to make this kinematic self-replication possible. Multiple generations were observed. The coverage was enormous. The word "robot" appeared in virtually every headline.

The "robot" label has real utility. It signals that these things are designed, intentional, and controllable in ways a wild organism isn't. It communicates that they have a purpose imposed from outside. And it keeps the public from assuming that scientists have spontaneously generated novel natural life. All of that is worth saying.

But here is what the word quietly smuggles past us. A robot is, by definition, not alive. It doesn't have to eat. It doesn't heal itself from genuine biological injury. It doesn't make copies of itself via cellular machinery. It doesn't respond to its environment with anything resembling autonomous behavior. Xenobots do all of those things. Calling them robots doesn't describe what they are. It describes a specific category they're supposed to stay in.

Section Two

What the Evidence Actually Shows

There are standard criteria for what biologists mean when they call something alive. They vary slightly by textbook, but the core list is fairly settled: metabolism, cellular structure, growth and development, response to stimuli, homeostasis, reproduction, and heritable information. Apply those criteria to a xenobot and something interesting happens.

The cells that make up a xenobot are not modified. No gene editing. No synthetic biology at the molecular level. These are normal frog embryonic cells, arranged differently. The skin cells that would normally cover a frog embryo's body sit on the outside of the xenobot and provide structural scaffolding. The heart muscle cells, which in a normal frog would pump blood, instead contract rhythmically and drive the bot forward through water. The AI did not reprogram the cells. It arranged them so their existing behaviors produced locomotion.

That distinction matters more than it seems. The xenobot is not executing a program written for it. It is expressing what its cells do naturally, in a configuration that produces a new collective behavior. When you cut a xenobot in half, it heals. Not because it was programmed to heal, but because living cells do what living cells do. No robot has ever healed itself from a cut. Materials don't do that. Biology does.

The self-replication result from 2021 is even harder to fit into the robot frame. The kinematic method the xenobots use, moving through an environment, collecting cellular material, and consolidating it into functional copies, has no known precedent in the history of life. Existing biological organisms replicate through cell division, sexual reproduction, budding, or sporulation. What xenobots do is genuinely novel. A robot that redesigns its own replication strategy in ways no existing machine or organism has ever used is not behaving like a robot. It is behaving like something biology hasn't made before.

The anthrobots arrived in late 2023, published in Advanced Science in 2024. Saudade Gumuskaya, working in Levin's lab at Tufts, derived them from adult human lung epithelial cells (tracheal cells obtained from donors) rather than frog embryo stem cells. Each anthrobot begins as a single adult cell. Over two weeks in culture, it self-constructs into a multicellular motile structure, between 30 and 500 microns wide, propelling itself using cilia. Nobody sculpts them. No external machinery gives them shape. The cells assemble themselves.

The anthrobots live 45 to 60 days. In one critical experiment, a sheet of human neural cells was scratched to create a wound. Anthrobots placed in contact with the damaged area stimulated repair: neurons re-grew across the gap within three days. The anthrobots weren't programmed to do this. They were placed near damaged tissue and their cells, human cells, did what cells sometimes do when they encounter injury. They helped.

And then in March 2026, the same team introduced neurobots: xenobot-like structures into which neural precursor cells had been implanted during construction. Those cells matured into neurons with functional axons and dendrites, forming primitive electrical networks inside the organism. Calcium imaging confirmed the neurons were active. The neurobots moved in more complex patterns than their non-neural counterparts. A living machine had grown its own nervous system. Nobody programmed that either.

Section Three

Why the Gap Exists

The "robot" framing didn't emerge from carelessness. It emerged from a genuine communication problem, one the researchers themselves helped create. When the 2020 paper arrived in press offices, it needed a sentence. "Scientists create first living, self-healing robots" travels further than "scientists create novel multicellular organisms by arranging embryonic frog cells according to AI-generated body plans." The second sentence is more accurate. The first sentence got published.

The researchers were not wrong to use the word. Bongard is a roboticist by training; his frame is legitimate. And calling these things organisms raises immediate and uncomfortable questions that "robots" neatly sidesteps. If they are organisms, are they individuals? Do they suffer? Can we build and destroy them at will? Can we patent them? Are they alive in a morally relevant sense? "Robot" keeps those questions at bay. It signals that this is a technology story, not a philosophy story.

But the philosophy story is exactly what's happening.

Michael Levin's broader project makes this clear. His lab at Tufts works on what he calls the bioelectric code: the idea that alongside the genetic sequence, cells carry and exchange electrical signals that collectively direct body-building. The genome says what proteins to make. The bioelectric layer tells a cluster of cells where to put everything, what shape to become, how many fingers to grow. Xenobots and anthrobots are, in this framing, demonstrations that biological cognition and decision-making are not confined to brains. They happen at the cell level. They happen in clusters of cells with no neurons at all.

Levin has put this plainly: his aim is to “dissolve the imaginary sharp line between a bot and a living organism, to show how every system offers aspects of both, and how not even ‘robots’ are the mechanical machines envisioned by our limited (and limiting) formal models.” That is not a hedge. That is a direct challenge to the framing most science coverage reached for without thinking.

The "robot" label also carries a specific assumption about agency. Robots execute instructions. Xenobots don't. They do what cells do, in a context that produces behaviors nobody wrote down. The anthrobot's neural repair behavior is the clearest example. Nobody told those cells to help injured neurons. Nobody designed that interaction. It emerged from what human cells naturally do when they encounter the right chemical environment. The AI designed a body plan. Life did the rest.

There's an institutional reason the framing has persisted, too. Funding for "bio-inspired robots" is easier to secure than funding for "novel living organisms with unclear moral status." Journals and press offices have learned which framings generate coverage and which generate letters from ethics boards. The gap between what xenobots are and how they are described is not accidental. It's the shape of every incentive in science communication working at once.

Section Four

What Accurate Framing Opens Up

None of this is an argument against xenobots. It's an argument for calling them what they are: living things, designed by AI, that don't fit our existing categories and shouldn't have to.

If you call something a robot, you ask robot questions. Can it be switched off? What is its range? Who is liable if it malfunctions? Those are real questions. But they're not the right first questions for something built entirely from living cells. The right first questions are biological. What are its interests, if any? What does it experience, if anything? Does cellular self-organization at this scale constitute any form of cognition? What happens when it's made from human cells and it starts doing things nobody designed it to do?

The anthrobot neural repair result crystallizes this. A structure made from human lung cells, assembled without any external form-giving, was placed near damaged human neurons and helped them heal. This has obvious therapeutic implications: patient-specific biobots derived from a person's own cells, introduced into the body to repair tissue, with no immune rejection because the cells are genetically identical to the patient's own. The appeal is real. But so is the question. When anthrobots are made from a specific person's cells and placed inside that person's body, what exactly is the relationship between the patient and the organism doing the healing? Are they separate? Are they extensions of the same biological individual?

Existing bioethical frameworks are not built for this. The 14-day rule governing human embryo research doesn't apply here because anthrobots don't come from embryos. Regulations on genetically modified organisms don't apply because no genes are modified. Regulations on medical devices don't apply because these things are not devices. They are alive, and the category is empty.

The neurobot adds a further layer. A living structure that has grown its own functional nervous system, that shows electrical activity in primitive neural networks, that moves in self-generated patterns: this is not the same object the 2020 xenobot was. It is a more complex entity. Whether complexity at the cellular level reaches any threshold of morally relevant experience is a question nobody can currently answer. But you can only ask it if you start by calling the thing alive.

This is Levin's deeper point, and it is not mysticism. It is a scientific claim about where cognition lives. His bioelectricity framework suggests that the capacity to process information, to integrate signals, to move toward goals, is not a property of brains alone. It's a property of organized biological matter at many scales. Xenobots are small and they live for weeks. But they navigate, they self-repair, they aggregate, and the newest versions have functional neural tissue they grew themselves. At some point on that trajectory, calling them robots becomes not just inaccurate but actively misleading about what is happening.

What's happening is that life is more plastic than we thought. A genome is not a blueprint; it is a toolkit. The cells in a frog embryo are not committed to making a frog. If you arrange them differently, they make something that has never existed. They still want to live. They still cooperate. They still heal. The AI did not make them alive. It made them into a new kind of alive, and we don't have a word for it yet.

“Robot” is the wrong word. “Organism” alone isn't quite right either. What we have is something genuinely new: a designed living entity, shaped by artificial intelligence, built from unmodified biological cells, behaving in ways nobody fully anticipated, and raising questions our ethical frameworks haven't caught up to. That is not a reason to keep the comfortable framing. It's a reason to let go of it.