The next Anduril won’t build missiles

Let’s play a game. Guess the category.

The customer is often the government. Sales cycles are painful. The incumbents are enormous. The technical talent lives in a different universe from the people writing the checks. Investors can explain, at considerable length, why venture capital doesn’t work here. Threats are getting worse. Technology is advancing almost too quickly. The industry connecting the two is barely moving.

Defense tech, circa 2014–2016.

Also, biodefense, circa…now.

That resemblance has been nagging me. Experts spent years explaining why startups couldn’t build serious defense companies. Then a few founders decided to build them anyway, and an entire category of investors discovered that they had always “believed in it since day one”.

I think biology is approaching the same turn.

Biology is becoming a defense domain. The tools for understanding and engineering living systems have taken transformative leaps, and bad actors get to use them too. The question is, who builds the companies that keep the defense moving as quickly as the threat. Who builds the Anduril for biodefense?

As if you needed something new to worry about this month after the deluge of AI existential dread, allow me to introduce you to the threat of bioweapons.

I’ve been pulling my thoughts together on this for a few weeks. Apparently, the world interpreted that as a deadline to provide more supporting evidence.

Anthropic’s threat intelligence report from a week or two ago describes five concerning examples of AI use in biological research. They include work involving viruses and paralytic toxins, state-associated research, and attempts to get around safeguards. Some AI assistance was blocked; some got through.

(Disclaimer - I had to build that table image by hand. Every AI image tool I tried refused⁵. One decided it might be against policy and stopped mid-task; another produced clip art and then asked whether I'd like to talk about something else. The illustration at the top of this piece is what survived.)

These are not five confirmed bioweapons programs. The report is explicit about uncertainty, dual-use research, and the limits of what the company could establish. But it puts something concrete under a debate that has long focused on hypothetical bad actors. People doing concerning biological work are already bringing AI into the process.

Then there is the September reporting on Russia’s newly refurbished biological research facility on Lisiy Island. RFE/RL’s investigation documents the construction and military connections around the site; Ukrainian intelligence alleges work relevant to biological weapons. Satellite images can show a new laboratory. They cannot tell us exactly what is happening inside it¹.

These are two different ends of the same threat. States can build deeper biological capabilities. AI can spread pieces of that capability to people who previously lacked the expertise. The scary change is not that a chatbot can suddenly manufacture the apocalypse - Physical biology (luckily) remains messy and difficult. But it is scary that bad actors can do more of the work, faster, while the defense is still organized around a short list of things it already knows to fear.

Discussions of biodefense collide with pandemic preparedness, especially post-COVID. There is certainly overlap - better diagnostics, manufacturing, and medical countermeasures help in both cases. But a pandemic doesn’t read your preparedness plan and look for the holes.

An adversary can.

The traditional defense world has just spent the last few years learning that you cannot sustainably fire a $3 million interceptor at a $30,000 drone. The cost asymmetry breaks: when offense gets cheap enough to proliferate, a defense built around bespoke, expensive responses eventually collapses.

Biodefense is wafting toward the same problem. The cost curve for designing, modifying, and testing novel biology is plummeting, while much of the “defensive” architecture still assumes bespoke detection, bespoke countermeasures, and bespoke programs for individual threats. An adversary’s objective might be to disrupt a military deployment, overwhelm a particular region, damage food & agriculture supplies, or just create enough uncertainty to paralyze decisions. It does not have to look like a movie about the end of civilization to be strategically useful².

Which gives the defense a different job. It has to keep working while someone is actively trying to make it fail, like a base that protects its people and continues its mission while the signal is still ambiguous.

A credible biodefense has three jobs:

The first two are deterrence by denial: If an attack is unlikely to create the panic or damage its sponsor wants, it becomes less attractive. Attribution adds another cost³. All of this together changes the game theory of whether the attack is worth attempting in the first place. A defense that is never used can still be the one doing the most work.

Find the threat. Try to disable the threat. Keep functioning. Figure out who was responsible in parallel.

Jeez, that sounds like defense because it is defense.

If I had to pick one number to organize this industry around, it would be time-to-countermeasure.

How long between recognizing a threat and getting an effective response into the hands of the people who need it?

Moderna went from selecting a sequence for its COVID vaccine to dosing the first participant in a clinical trial in 63 days. That was extraordinary. It was also the beginning of human testing. The rest of the journey still had to happen.

CEPI’s 100 Days Mission aims to compress the path from recognizing a pandemic threat to a vaccine ready for initial authorization and manufacturing at scale. It is a serious ambition, and a useful way to frame an entire system to work backward from a deadline.

An adversary does not have to give us 100 days.

So the response has to include what is available before a new vaccine arrives: detection, protective measures, existing treatments where appropriate, and the ability to keep critical operations running. Some defenses should already be in place. Others need to be updated quickly as we learn what we’re dealing with.

The threat does not pause while the response moves from one political institution to the next.

This is why time-to-countermeasure is more than a KPI. I think it dictates a new company architecture.

You can have a great piece of biosurveillance hardware, a great bioanalytical model, and a great manufacturing process, yet still have a terrible defense system if nobody takes radical ownership of the movement between them. 

A decade of defense investing has taught me the difference between selling a component and owning the mission. The component can be valuable. The company accountable for the outcome usually defines the category. Biodefense has plenty of promising components. It has almost nobody owning the response.

String those handoffs together, and you get a loop: Detect something unusual. Characterize it well enough to act. Select or design a response. Test it. Manufacture it. Deploy it. Feed what happens back into the system so the next response improves.

None of those steps is new. What's new is that sequencing, computational design, lab automation, and increasingly flexible biomanufacturing have all gotten dramatically better at roughly the same time. For the first time, you could plausibly run the whole loop fast. The same convergence is what made the threat cheaper.

They just don't connect very well. Each step sits in its own world, with its own institutions, budgets, data formats, and people who will tell you the next step belongs to someone else.

That creates two kinds of company:

A sensor alone is not an air-defense system, and a good vaccine alone is not a biodefense capability. Integration is a large part of the work. 

To be fair, stockpiles of masks and drugs to fight known threats still matter. But a warehouse full of answers to yesterday’s questions cannot be the entire strategy when someone is working on tomorrow’s. The defense has to generalize faster than the threat can proliferate.

So why hasn’t anyone built the loop? It is certainly not because nobody noticed the problem.

The United States has no shortage of institutions, plans, or good people. Johns Hopkins' health-security center dates to 1998. George Mason runs a master's program in biodefense. The Bipartisan Commission on Biodefense's Apollo Program has an ambitious technology agenda, with a July update adding AI as a priority. BARDA has funded real countermeasures, and there are people alive because that work exists.

An academic program produces experts. A commission produces recommendations and white papers galore. A grant can occasionally produce a specific breakthrough. None of them produces an organization that can answer the phone at 2 a.m. and deliver an updated defense.

The people inside these institutions understand the threat perfectly well, and some have been trying to industrialize the response for years. The challenge is that the incentives around these places are still better at rewarding the white paper than building and maintaining an industrial capability.

We need less conference-panel energy. More factory acceptance test.

There is an inconvenient step between “civilization needs this” and “this is a venture business”: somebody has to buy it.

More importantly, what the government chooses to buy will determine what kind of companies get built. The defense-tech founders who broke through did not make procurement disappear; they found a way through it. Biodefense founders will need to do the same, and the government will need to decide what it is actually purchasing.

A research project? A batch of doses and masks? Or a standing ability to identify a threat and respond within a specified time?

I’d argue that more of the budget needs to buy option three. Pay for readiness: monitored sites, regularly tested response capability, maintained manufacturing capacity, and a team accountable for performance. Exercise it before the emergency. Measure the time it takes. Find the broken handoffs while the consequence is an embarrassing test result.

Keeping capacity ready costs money even when nothing happens. That is a normal feature of defense. We do not conclude that an air-defense battery had a bad quarter because nobody fired a missile at it.

The founder’s job is to turn the broad mission into a first product somebody can buy. Protect a defined set of facilities, shorten one part of the response by an order of magnitude, make a capability reliable enough that an operator can depend on it. Then earn the right to expand.

Commercial work can help pay the bills and improve the technology between government contracts. It can also consume the entire company. If the urgent customer is always a pharmaceutical company and the defense mission is always on next year’s roadmap, you may have built a perfectly good biotech supplier. You have not yet built the company this note is about.

I would want to know who owns the budget, what they will buy twice, and which part of the response gets measurably faster. Strategic importance does not answer those questions for you.

It does make them worth doing the work on.

The field is small enough that individual companies still tell you quite a lot about where it might go.

Valthos and Red Queen Bio ⁴are pursuing the use of AI to recognize biological threats and develop countermeasures much faster. Pilgrim is producing the Argus system, which will try to deliver continuous biosurveillance of airborne pathogens. They look less like traditional biotech companies waiting for a defense application, and more like the first companies being built around biodefense itself.

These companies are early. But they are evidence that founders are beginning to organize around the problem, and that some investors are willing to pay for more than a policy recommendation. Perhaps not surprisingly, in these early days, these ventures seem to be targeting the “handoffs” rather than “owning the full loop”. That is coming next.

My confidence in this sequence of events stems in part from some of the work we’ve been backing at ACME at the other end of the stack. My partner Christian and I have argued for years that biological industries are approaching a strange transition point as advances in bioprocess data, AI, automation, and biomanufacturing begin to converge. Some companies we’ve backed like Invert, Reshape, and Again are attacking hard problems across those layers.

These are very different businesses from the ones I’m describing in this note, and I’m not retroactively handing them all “biodefense Eagle Scout badges”. But they are helping make biology more measurable, repeatable, and industrial. These are some of the “picks and shovels” underneath whatever new capabilities get built on top.

We’re convinced that convergence will produce business models that barely existed before. Biodefense is one of the most interesting (and most urgent) places we think that happens.

Go back to the opening.

Huge incumbents. Difficult government customers. Technical teams that struggle to meet the people with budgets. Investors with very sensible reasons to stay away.

A decade ago, that description helped keep people out of defense tech. Today, founders and investors are competing to build the next drone, the next counter-drone system, and the next counter to the counter-drone system. It is, of course, useful work. It’s also an increasingly crowded room.

There is something very venture-capital about waiting until a market has produced giant companies before deciding it might produce giant companies.

The early defense-tech founders and investors were not right because procurement was secretly easy. They were right that the threat, the technology, and the kinds of companies capable of connecting them were changing. The old objections described the market they had inherited. They did not settle what could be built next.

That is the pattern I see in biodefense.

The analogy will not do the biology for us, but it should make us suspicious of the argument that an industry’s current structure is its permanent one.

We know how to recognize this movie now. We should be getting better at showing up before the credits.

Biology is becoming a defense domain. The technical stack is arriving, and the company ecosystem hasn't caught up. The defining companies will be built around a simple, unforgiving promise: when the threat changes, the defense changes faster.

— AF