DARPA Wants to Write DNA With Light Inside Living Cells. It's Not Mind Control — and Biologists Aren't Sure It's Possible.

Over the past few months a DARPA program called Generative Optogenetics has been described, depending on the website, as a plan for “covert biological manipulation,” a step toward “neuro-weapons,” and proof of a “Luciferian” agenda to rewrite the human genome with light. On a different set of websites it is the dawn of programmable life — “the next major thing to happen after AI.” Both camps are reacting to the same document, and neither has read it closely.

Here is what DARPA actually published: a 40-page program solicitation, number DARPA-PS-26-10, asking research teams to design a protein complex that lives inside a cell and builds DNA or RNA to order, using pulses of light as the only input. DARPA calls this hypothetical machine a “nucleic acid compiler.” It does not exist. As of this writing, the program has not funded a single team — and the list of who even applied is not public.

What a “nucleic acid compiler” is supposed to do

The pitch is genuinely radical, and worth stating precisely because almost no coverage does. Every method we currently have for getting genetic instructions into a cell — viral vectors, lipid nanoparticles, electroporation — works by physically shoving a molecule of DNA or RNA across the cell membrane. The solicitation’s central observation is that this is the bottleneck.

GO proposes to take the molecule off the trip entirely. Instead of delivering DNA, you deliver information, encoded in light. A protein complex expressed inside the cell — the compiler — would read pulses of different wavelengths, map them to the four genetic bases, and assemble the corresponding strand from the loose nucleotides the cell already keeps on hand. Light in, gene out. The programming metaphor is DARPA’s own: the document describes “a direct interface between computers used to design genetic sequences and living cells that operate on those sequences.”

The targets are aggressive. By the end of the 42-month program a winning compiler is supposed to synthesize a 3,000-base sequence inside a living cell, at roughly one base per second, with fewer than one error per million bases — and to write two different sequences into the same cell within an hour of each other. The advertised applications run from in-place medicine (reprogram cells without needles or nanoparticles) to agriculture and biomanufacturing. The hook DARPA leads with is spaceflight: rather than haul a pharmacy to Mars, carry a light source and a data file and synthesize the therapeutic inside the astronaut on arrival, “diminishing reliance on brittle supply networks that become untenable for long distance operations.”

Why this is not mind control

Most of the alarmed coverage makes the same mistake, and it is an understandable one: it hears “optogenetics” and reaches for the famous version. Classical optogenetics — the field that genuinely dates to around 2005 — puts light-sensitive ion channels like channelrhodopsin into neurons, so that a flash of light makes them fire. That is the technology behind the lab experiments where a laser steers a mouse’s behavior, and it is why the word “optogenetics” carries a whiff of remote control.

Generative Optogenetics borrows the same underlying trick — a protein that changes shape when light hits it — and points it at a completely different job. Here the light-sensitive domains don’t fire neurons; they gate an enzyme that builds nucleic acids. The output isn’t a nerve impulse, it’s a strand of DNA. So when a write-up slides from “light controls cells” to “implant false memories and hallucinations” — as one tech outlet’s coverage did, importing imagery from neural-optogenetics work in mice — it is borrowing the reputation of one technology and pasting it onto a different one that happens to share a word.

That doesn’t mean there is nothing to worry about. It means the real concern is somewhere else, and we’ll get to it.

The biologists are not impressed

When the program reached Hacker News, the most useful reaction wasn’t fear — it was a shrug from people who do this work. The sharpest comment questioned the entire premise of doing the synthesis inside a living cell at all:

The objection is substantive. We can already synthesize DNA outside cells, and it is hard: today’s enzymatic methods run at minutes per base with error rates between roughly 1 and 20 percent per base, and the chemical gold standard caps out near 200 bases before fragments have to be stitched together. GO is asking for one base per second, one error in a million, thousands of bases long — and to do all of it inside a living cell, a far more crowded and hostile environment than a test tube. Attempting the hard thing in the hardest possible place is the program, and at least some practitioners think the numbers are fiction.

The other recurring note was that the idea isn’t new. Commenters traced “optical-to-DNA” proposals back through 2017 and 2011, and one well-known figure in the DIY-biology world said he had watched the concept circulate among “almost every major household name lab PI” since 2008. The field of optogenetics itself, another pointed out, has been established since 2005. The interesting question is not whether someone thought of this before; it’s what DARPA believes has changed to make it fundable now.

Notably, the program’s own leadership turned up to take questions. A commenter posting as Michael Koeris, identifying himself as director of DARPA’s Biological Technologies Office, offered to host “an AMA style explainer and engagement.” That is not the behavior of an agency running a covert program — it’s an agency trying to recruit a research community for a long-shot one.

The dual-use problem DARPA is already worried about

If there is a real reason for unease, it isn’t neurological — it’s that a light-triggered machine that writes arbitrary DNA inside living cells is, by construction, the sort of thing biosecurity people lose sleep over. DARPA plainly knows this. Before funding a single research team, it stood up three advisory bodies: a Biosecurity Working Group, a Regulatory Policy Working Group, and an independent commercialization group. The biosecurity panel is charged with running tabletop exercises and coordinating with the EPA, FDA, and USDA to map out “misuse, accidental release, or malicious exploitation.”

The solicitation draws its own hard lines: no work in human embryonic stem cells, no export-controlled or biosafety-restricted genes, and any functional compiler’s output is to be classified as Controlled Unclassified Information because, in DARPA’s words, it “will have similar metrics to export controlled synthesizers.” That last point is the tell. The agency is treating a successful compiler as the equivalent of a regulated DNA-synthesis machine — which is exactly what it would be, just running on photons inside a cell instead of reagents in a benchtop box. That is the legitimate dual-use story, and it’s more interesting than mind control precisely because DARPA is funding the capability and pre-litigating its dangers in the same document.