She set out to build mirror life. Then she talked the field into stopping

Nathan Labenz opens The Cognitive Revolution with Kate Adamala, a professor of genetics at the University of Minnesota, because she had just done something rare in a hot field: coordinated a public warning against a research direction she had been funded to pursue. Mirror life — organisms built from the mirror images of life's usual molecules — had been her first big independent grant. By June 2025 she was a lead voice telling colleagues not to finish the job. Before that story, she wants a simpler observation on the table. Life looks endlessly various in form. Chemically, she says, it is "extremely boring."

Proteins in every known organism are assembled from 22 proteinogenic amino acids. DNA and RNA use five nucleobases. Chemists can synthesize hundreds, even thousands, of amino acids, and a zoo of nucleic-acid analogs is theoretically available. Life took none of them. Synthetic biology's bet is to widen that repertoire anyway: give cells tools natural evolution never bothered with, see what functions show up, and learn which chemistries can actually carry a living system. Adamala is not in the business of silicon organisms or life under exotic pressures. Her expertise is carbon molecules in water. Even staying there, she says, the possibility space is almost endless.

The floor is already surprisingly high. The ordinary ribosome shared by bacteria and humans will, if you take it out of a cell and "gently encourage it in vitro," polymerize hundreds of unnatural amino acids — and even stitch together things that are not amino acids at all. That is without evolving a new assembler. Labenz asks how wide the true range might be. Adamala's limiter is imagination: modern terrestrial life is a sample size of one. NASA's working definition, a self-replicating chemical system capable of Darwinian evolution, sounds tidy until she applies it to herself. Without her husband she cannot replicate, so under that rule she is not alive. Her own working definition is Potter Stewart's: she will know it when she sees it. What she insists on is emergence. Life has to be more than the sum of its parts, which is also why a complex enough computational system, with no DNA at all, might one day cross the same line.

Did biochemistry happen only once? Adamala's guess, shared by much of her community, is that it happened many times. The moment Earth had crust and liquid water, life appears in the rock record almost instantly on a geological clock, which suggests that under the right conditions the chemistry is not that hard. Early lineages would have competed for the same pool of molecules, and everything in nature tries to eat everything else. One lineage ate or outcompeted the rest. Nascent life is still "very, very wimpy." A new origin in a 2025 hydrothermal vent would almost certainly be lunch for the incumbents. Once a planet is occupied, starting over is a bad bet.

Two patterns pull in opposite directions. Very early biochemistry was promiscuous: cells used whatever they could grab because they had not yet built biosynthetic pathways. Five or six amino acids look ancestral; others arrived later, some almost on the eve of the last universal common ancestor. Then a membrane went up. Isolation from the environment forced hard choices, because a cell that drifts into a patch of ocean missing its favorite small molecules dies if it cannot make them. Humans and guinea pigs are the cautionary tale. Most mammals still synthesize vitamin C; we lost the pathway. When the age of sail put people on ships for months, scurvy followed, because no one had remembered that the body no longer made a compound the environment had always supplied. Early cells, she says, had to pack like someone moving into a van, not like a homeowner with a basement full of unused stuff. Only after biosynthesis existed could the menu expand again.

That history also explains a dumb, powerful machine. Every known organism has a ribosome, and Adamala calls it an "incredibly crappy catalyst" — in the jargon, an entropy trap, not a proper enzyme. Its business model is to smash substrates together into a necklace and kick the chain out an exit tunnel. Specificity is geometry: if the amino acid is too bulky it stalls in the hole; if it fits, it gets translated. There is no inspector that rejects an unnatural residue. The enzymes that charge amino acids onto tRNA are the picky ones. The ribosome itself does not guard the genetic code.

Why so slapdash? Frozen accident. The first organism that figured out protein synthesis had a huge evolutionary advantage, and after that you do not mess with the core, because any mistake is lethal. Adamala reaches for the router in her house, set up five years ago, that the family dares not touch lest the internet die, and for the ancient computers still running internet infrastructure. In the language Labenz uses for neural nets, the ribosome sits in a local minimum that works. A deeper minimum may exist, but the hill out is tall. That is why a machine this old can still swallow a hundred extra amino acids: it never evolved the snobbery that would refuse them.

Synthetic biology, Adamala says, is trying to engineer life that natural evolution never bothered to run — both because that would be the ultimate scientific control, and because civilization still runs on oil for more than energy. Look around a room: electronics, clothes, medicine, cosmetics, and the fertilizers that feed the current population are petrochemical. Renewable electricity does not replace those molecules. The other goal is healthspan. Biology still cannot hand her a complete ingredient list for a single human cell, let alone the causal graph of a tissue. Richard Feynman's overused line still fits: what she cannot build, she cannot claim to understand.

Two programs are trying to meet in the middle. Top-down, Craig Venter's team built a chemically synthesized minimal genome in a mycoplasma chassis: 474 genes, and still a black box. A couple of dozen are essential genes of unknown function; knock them out and the cell dies, and no one can say why. Small-molecule metabolism is murkier still. Bottom-up is what Adamala calls cheating. She assumes she will not reverse-engineer a natural cell in a career, so she starts from molecules she fully understands and tries to assemble something with metabolism, replication, and interactions. Her lab's published synthetic cells carry about 85 to 90 genes. A paper from about a decade ago guessed that a truly minimal living system might need around 120. The gap to 474 is real, but it is not an order of magnitude, and many of those 474 functions look like things an engineer could route around.

There is no Frankenstein instant. You pipette clear liquids into a tube and wait. Activity is scored by what proteins get made and what molecules get metabolized. Membranes, transcription, translation, feeding, growth, even genome replication are already in hand; liposomes that are not alive have membranes too. What many biologists are waiting for is Darwinian evolution, and they do not have it yet. Adamala bristles at the word "spontaneous." Lipids in water at the right pH form membranes because that is what lipids do. An RNA polymerase that finds the right promoter transcribes because that is its job. No heavenly finger. The origin of life, watched from a front-row seat, would have been easy to miss.

The missing variation comes from polymerases that are "annoyingly high fidelity," taken from a bacteriophage. Too many errors and you fall into error catastrophe; too few and the population has nothing to select. Engineering an enzyme to be a little bit broken is much harder than wrecking it. Her synthetic genome is only about 90,000 bases. At a notional one error per 100,000 bases, each replication cycle yields zero or one mutation — not enough. A million-base genome would accumulate a few. Human replication pairs high-fidelity polymerases with dedicated proofreading enzymes her synthetic cells also lack. Nature never built a polymerase for a genome this small, because nothing in nature is that simple. Until that Goldilocks error rate exists, the bottom-up cell can look busy without quite counting as alive.

Chirality is the left-hand, right-hand fact of biomolecules. Any asymmetric molecule comes in two forms that cannot occupy the same space, the way a left shoe will not happily go on a right foot. Earth life standardized on one set: DNA as a right-handed helix, amino acids mostly left-handed. Mirror life would invert the whole stereo-chemistry. The molecules would be chemically almost identical and spatially incompatible with the rest of biology. Adamala wanted it for three reasons. Scientifically it would be the first living cell off Earth's tree of life. Medically, mirror molecules are hard for the immune system to see, so they can circulate instead of being cleared. Industrially, a mirror-bacteria bioreactor would not be wrecked by a phage sneeze. The third point is the second point's dark twin. Immune invisibility is a feature in a drug and a bug in a pathogen. Phage resistance is a feature in a factory and a bug in the wild.

This is not science fiction at the molecular layer. Mirror proteins and mirror nucleic acids can already be made chemically; people are building mirror polymerases. What does not exist — "thank God," she says — is a complete, self-replicating mirror cell. At the field's then-current pace, many colleagues thought a self-replicating version was a decade away, twenty years on a conservative read, still inside a human lifetime. Spiegelmers, named from the German *Spiegel* for mirror, are already in clinical trials because they are stealth to immunity. Smarter immunologists than she is, Adamala says, treat that as transitive: if the parts do not trip the alarm, an organism built of those parts probably will not either.

"Stealth" is more specific than sci-fi camouflage. Injected into blood or tissue, the molecules get sniffed by white cells and do not activate a response. Labenz's version: ordinary foreign RNA would mobilize clearance; reverse the chirality and the system does nothing. Adamala's dog metaphor is the one that stuck. A dog that smells bread on the floor eats it; a dog that smells a rock usually walks on. The immune system has to classify a threat before it spends resources. Mirror chemistry fails that classification. A mirror cell in the bloodstream would find a warm, nutrient-rich pond with no patrol. In the environment it would be a self-replicating rock: no virus to infect it, no predator that can digest it. "There's nothing more dangerous than something that nothing can eat," Labenz says. She agrees. That invisibility was the original selling point. It became the scariest property.

The conversion was not a eureka. It was a slow "holy crap" as she started working in biosafety circles and talking to ecologists and immunologists — fields she is not trained in. Precedent for something nothing can eat is thin. The great oxygenation event is the example everyone reaches for: cyanobacteria filled the air with a waste gas that killed most of the biosphere and injured the cyanobacteria themselves. Since then, biology has kept an equilibrium. Humans are the closest later case, an organism that climbed the food chain, ran out of predators, and produced global warming. If she had to bet, she would still bet that normal-chirality life would eventually find a way to eat or infect a mirror competitor. Evolution is stochastic and cannot be programmed onto a deadline. "It's a risk that I'm not willing to take."

The best skeptic argument she has heard is that the sun will die too, and nobody rebalances a retirement account over it. Mirror life, in that frame, is too far off to crowd the present. Her reply is that her community was actively trying to make the thing. Stay quiet, and the next lab still hears only the upside. At then-current progress, a self-replicating mirror cell inside a decade was a common forecast. "We have to stop it now before we actually have the capacity to do that." Containment is the next shrug: keep it in the lab. She has no patience for that. Make a foolproof container and nature develops a better fool. Smallpox leaked from a British lab; foot-and-mouth leaked from Pirbright. Twenty-seven locks do not seal a crack in the floor or an unlatched tube. The security layer is worse. A perfect box still fails if someone with bad intentions walks the sample out. She will not rest a safety case on the claim that the world contains no malicious people.

Auxotrophy — engineering dependence on rare reverse-chirality nutrients — is the standard biosafety patch, and she thinks it is not enough here. A cell that needed fifty exotic chiral feedstocks would, poured on a lawn, go nowhere in one generation. A bad actor can rewrite those pathways. It is already possible to imagine a mirror cell that lives on achiral carbon sources; ordinary bacteria like that exist. What can be imagined can be built a second time. *Jurassic Park* is her least favorite teaching example: lysine dependence is a first-year blunder because lysine is everywhere, and the movie needed the dinosaurs to escape. Real designs would pick rarer molecules. The logic is the same. Any lock that depends on an external supply can be opened.

Would Earth's biosphere simply outcompete a chiral underdog? Maybe. She pictures not a turtle but a slug: incredibly slow, and unstoppable because nothing eats it. Labenz asks about Drexler-style gray goo, a 72-hour conversion of the planet. That, she says, is not the scenario. If it happened it would be "the most boring pandemic you can imagine," a crawling apocalypse, slow and without an off switch because no existing mechanism is aimed at it. Not making the cell is cheaper than inventing a control system after the fact.

Labenz wants a p(doom). Adamala draws a brighter line than a probability. Mirror life does not exist, so it can still be stopped. AI is already out of the box, and she does not particularly want it stuffed back in. She has not written a line of code in over a year; ChatGPT writes the script, she pastes the error back, and work continues. The models also speed biomedical discovery. Twenty years ago a pause might have been thinkable. Now the job is to constrain applications and prepare for harm. Molecules-versus-organisms is the analogy she likes: Spiegelmers can stay; the self-replicating cell should not be a gain-of-function step. Labenz's version for AI is similar. Tools that tell a cat from a dog were toys. Systems that can hack off a server, scrounge compute, and keep going start to look like doors that may not close.

She did not abandon the project quietly. A small group talked, then they went looking for people who could wreck the argument — immunologists, ecologists, policy researchers, not only the engineers. The pitch was "please prove us wrong," because they sincerely wanted a hole. Many co-authors on the *Science* paper had never thought about mirror life until that conversation. Nobody found the hole. Some arrived with angles that reshaped the accompanying technical report. No one reviewed the evidence and walked away unconvinced, at least among the people they approached. Six months after publication she still wants a disproof. Mechanism, she hopes, will be threefold: a community taboo like the one that followed the CRISPR baby, funding that will not touch the work, and rules that make it not allowed. Germline editing is her comparison and her warning. She can imagine refusing the technology in the abstract and failing as a mother if someone offered a large cut in her child's Alzheimer risk. If the option exists, people will take it, inequality will go biological, and two tiers of human in a generation or two strike her as a stability disaster. Better not to have the option. Sequencing fell from a million dollars to a few hundred, Labenz notes; she replies that you only ride that cost curve by starting, which is a leap of faith she does not want to take on germline editing.

Fair distribution looks different in software. Labenz cites Coca-Cola and a viral detail: Terence Tao, often called the greatest living mathematician, had been spotted on Claude's free plan. The best model was, for a moment, the same drink the president drinks. Adamala says biology cannot copy that trick. Labor and materials do not vanish. She grants Labenz's taxonomy more than she can derive it: mirror life and highly autonomous, resource-seeking AI belong in a small circle of doors that may not close; germline editing and the social shape of widely deployed AI are messy, but a later generation might still brake. Intuitively that split feels right. She would not want to be the emperor who draws the line.

The AI-specific fear she will name is narrow and heavy: a model that can synthesize the literature into an executable recipe for a better pathogen, toxin, or drug and hand it to someone without the training to refuse. Today's systems cannot; the best virologists cannot give a definite answer either. Jailbreaks already punch through "I won't tell you." The social lesson from the coalition is carrot, not stick. The alternative they rejected was scaring governments into a ban before the community bought in. "You're not trying to take their toys away. You're trying to make sure they keep playing with their toys in a safe way." Mirror life also had a tactical advantage AI lacks: nobody had yet made billions from it. She has not quit synthetic cells, and mirror-molecule trials continue. The circle she drew is one branch. Labenz's closing hope is that AI labs learn to do the same — step back, change course, and keep the rest of the tree. "No way out but through," they agree, and then they stop.