AI just invented a cure for snakebite. Then it invented a problem.

In partnership with

Get bitten by a black mamba in the wrong part of the world and your odds come down to a 19th-century invention: antivenom made by injecting venom into a horse and harvesting its blood. It’s expensive, it can trigger a dangerous immune reaction, and the right vial is often hundreds of miles from the person who needs it. Snakebite still kills somewhere between 80,000 and 140,000 people a year, by the World Health Organization’s count — most of them poor, rural, and far from a hospital.

In January 2025, a lab in Seattle published a different answer. They didn’t find a better antivenom in nature. They designed one from scratch on a computer — a protein that has never existed in any living thing — told an AI exactly which part of the venom to grab onto, printed the result, and watched it save mice from a lethal dose. Survival rates: 80% to 100%.

This week: the astonishing arrival of AI that invents the machinery of life — what it can already do, the example that sounds like science fiction, and the very sharp catch that a team at Microsoft went looking for on purpose.

🧬 What actually changed

Proteins are the tiny machines that run everything alive — they digest your food, fire your neurons, fight off infections. Each one is a chain of building blocks folded into a precise 3-D shape, and for decades the dream was simple to state and nearly impossible to do: design a brand-new protein with a shape you choose, to do a job you pick. Nature spent four billion years searching the possibilities. Humans, working by hand, barely scratched it.

Then AI cracked the folding problem, and went further. There’s an important difference here. AlphaFold predicts the shape of a protein that already exists. The newer tools create. The standout is RFdiffusion, out of the Institute for Protein Design at the University of Washington, led by David Baker — who shared the 2024 Nobel Prize in Chemistry for this work. You hand it a target and it dreams up a protein, atom by atom, that has never appeared in nature.

The leap isn’t just that it works. It’s the hit rate. Baker’s verdict on the snake-venom project says it all: the design software is now so good that they only needed to test a few molecules — instead of grinding through thousands in the lab. Design has quietly turned from a lottery into something closer to engineering.

💉 The cool part: medicine that gets designed, not discovered

The antivenom is the headline, but it’s one of several.

The snake-venom work, published in Nature, went after a nasty family called three-finger toxins — the fast-acting nerve poisons that classic antivenoms struggle to neutralize. The designed proteins are small, ferociously stable (they survive heat that would wreck a normal antibody — which matters a lot if your supply chain is a motorbike in rural India), and they protected mice against lethal doses. A shelf-stable, cheap-to-brew snakebite treatment would be a quiet revolution for the very people the current system fails.

Then came the bigger one. In late 2025, the same lab showed it could design antibodies from scratch — the Y-shaped guided missiles your immune system uses, and the basis of a $200-billion drug industry covering cancer, autoimmune disease, and more. Until now, making a new antibody meant immunizing an animal and screening enormous libraries, hoping something stuck. The new approach generates a full-length antibody on a computer, aimed at a precise spot on a target you specify. It’s early and success rates are still modest — but “design an antibody to order” was, until very recently, simply not a sentence you could say.

The throughline: for the whole history of medicine, we found our drugs — in molds, in soil, in tree bark, in horse blood. We’re now starting to design them. That’s the amazing part. Here’s the catch.

Turn AI into Your Income Engine

Ready to transform artificial intelligence from a buzzword into your personal revenue generator?

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

• A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential

• Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background

• Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get Your Guide

⚠️ …and the very sharp double edge

A tool that designs a protein to neutralize a toxin is, structurally, a tool that can design a protein that is one. Same software, opposite intent.

Here’s the part that should make you sit up. The thing standing between a bad actor and a dangerous protein isn’t really the AI — it’s the DNA synthesis companies. To turn a digital protein design into a physical molecule, you order the genetic code from a synthesis provider, and the responsible ones screen every order against databases of known toxins and pathogens and refuse anything that matches. That screening is the chokepoint that keeps the whole system safe.

In 2023, Eric Horvitz at Microsoft asked an uncomfortable question: what if generative design tools can rewrite a toxin so it still works — but no longer matches anything in the screening database? His team checked. Using open-source design tools, they generated roughly 76,000 reengineered variants of 72 dangerous proteins, including ricin. The variants kept the toxic function but looked, to the screening software, like strangers. Many sailed straight through the biosecurity filters meant to catch exactly this. They’d found the first biological “zero-day” — a term lifted straight from cybersecurity for a hole nobody knew was there. Published in Science in October 2025.

That’s the double edge of every powerful general tool, stated as plainly as it gets. The capability that designs a snakebite cure in an afternoon is the same capability that designs a screening-evading toxin in an afternoon. The machine doesn’t care which one you ask for.

🛡️ How the responsible people are actually handling it

The encouraging part: the people who found the hole are also the reason it’s mostly patched. And the playbook they used is one any field racing ahead with powerful AI should copy.

1. Treat it like a cyber vulnerability, not a press release. Microsoft’s team didn’t dump the findings online. They borrowed responsible disclosure wholesale — kept the dangerous specifics secret, alerted the people who could fix it, and waited to publish until defenses were in place. A “biological zero-day” handled exactly like a software one.

2. Fix the chokepoint, not the AI. You can’t un-invent RFdiffusion, and you wouldn’t want to — it’s curing diseases. So the patch went where the leverage is: the DNA-synthesis screening software. The team built and distributed upgrades that catch far more AI-reworked toxins, pushed out to synthesis providers before the paper dropped.

3. Be honest that it’s not solved. Even after patching, about 3% of the dangerous sequences still slipped through. The fix narrowed the hole; it didn’t close it. Biosecurity is now a patch-and-repatch cycle, just like infosec.

4. Keep the layers. No single screen is enough. Customer vetting, anomaly detection, controlled access to the most capable tools, international coordination — defense in depth, because any one layer will eventually be evaded.

5. Red-team your own miracle. The reason this story has a happy-ish ending is that defenders went looking for the dark use of their own technology before an attacker did. Assume your wonderful tool is also a weapon, and go stress-test it yourself.

The takeaway

A year ago, “AI designs a working medicine from scratch” sounded like a grant proposal. Today it’s a snakebite antidote that saved mice, an antibody industry quietly bracing for reinvention, and a Nobel Prize already in the cabinet. This is AI doing something deep and physical and good — not writing emails, but inventing the machinery of life.

And it’s the cleanest example yet of the rule that governs all of this: the technology is neutral; the intent is not. The same design tool points at a cure or a toxin depending only on who’s holding it and what they type. That’s not a reason to slam the brakes — the snakebite victims and cancer patients are real. It’s a reason to build the guardrails as fast as we build the capability: screen the chokepoints, disclose like grown-ups, and never let the wonder outrun the watchfulness.

We taught machines to design life. The hard part was never going to be the designing. It’s making sure we only like what they build.

Reply and tell us: as AI starts designing medicines, where should the hard line sit — the design software, the DNA-synthesis order, or somewhere else? Best answers get featured next week.

— itscybernews · written by a human, edited by a slightly nervous agent ·