Dear Decision Maker,

First, an apology.

Last week's newsletter didn't reach everyone.

Some of you got it, some of you didn't, and that's on us.

We had a technical issue on the send, and we've fixed it.

If you felt like you'd been dropped, you hadn't. Thank you for sticking with me.

Now, to this week.

I've been thinking about this topic for a while.

It keeps coming up:

• In client conversations
• In the headlines
• In the quiet questions people ask me when the recording stops

I've written before about AI in intelligence and in military operations.

Today I want to talk about something that scares people more, and I think for good reason.

AI in biomedical research.

Or to put it plainly: Using AI to help design new viruses.

Let me be careful with my words here, because this is a topic where fear runs ahead of fact.

Start with the ground truth

Right now, in the eastern Democratic Republic of Congo, there is an Ebola outbreak.

Not the strain you've heard of. This one is Bundibugyo.

It's only the third time in recorded history this strain has appeared.

There is no approved vaccine for it, and no specific treatment.

The tools we usually reach for were built for a different strain, Zaire Virus, and in the early days the diagnostic tests were coming back negative even when people were sick, because they were looking for the wrong virus.

As of late July, the numbers are more than 1,000 dead and over 2,500 confirmed cases.

The WHO is calling it the fastest-spreading Ebola outbreak on record. It hit 1,000 cases in about 40 days.

The 2018 North Kivu outbreak took the better part of a year to get there.

Why is it moving so fast?

Ground truth:

1. Armed groups control the territory
2. Populations are displaced
3. Health workers are being attacked
4. Communities are denying the virus exists and evading quarantine
5. The health infrastructure was thin to begin with

I am not telling you Ebola was made in a lab.

It wasn't. Nature wrote this one.

I'm telling you something else.

This is what an outbreak actually does in a country with a weak health system, no vaccine, and armed men at the roadblocks.

Remember that picture. Hold it in your head.

Because the next part is about how much easier it is becoming to write a virus on purpose.

The case study

In September last year, researchers at Stanford and the Arc Institute in California published something that should be on every decision maker's radar.

They used an AI model called Evo.

Think of it as a large language model, the same basic idea as ChatGPT, except instead of being trained on human text, it was trained on the genomes of roughly two million bacteriophages.

A bacteriophage is a virus that infects bacteria.

They pointed it at a small, well-understood phage: 11 genes, about 5,000 letters of DNA.

They asked the AI to write new versions. It proposed 302 designs.

They printed those designs as real DNA and tested them against E. coli.

Sixteen of them worked. Living, replicating viruses.

Designed by a machine. Killing bacteria in a dish.

The AI didn't just copy what it had seen. It invented viruses with new genes, truncated genes, gene orders that don't exist in nature.

Now, to their credit, the team drew a hard line. They deliberately did not train the model on any virus that can infect humans.

As one of the field's founders, Craig Venter, put it:

That line, "if someone did this," is the whole newsletter.

Possible is not the same as probable

Here is where I want to slow down, because this is exactly the kind of story where legacy media picks the scariest frame and runs.

We don't assess threats by how frightening they sound. We assess them across three things: capability, intent, and opportunity.

You need all three for a real threat.

Take one away and the scenario collapses into a headline.

Capability: Is it technically possible for someone to design a dangerous novel pathogen? The honest answer is that the door has opened wider than it was two years ago. AI lowers the expertise you need. Gene synthesis can be ordered. A modest lab and cheap equipment can do things that used to require a state programme. So yes: more possible than it was.

Intent: Who actually wants to do this? This is where fear and reality separate. Building a weapon that spreads without control is as dangerous to its maker as to the target. Most state actors know this. Most criminal actors want money, not an uncontrollable plague. The pool of people with genuine intent to cause mass biological harm is small. Not zero. Small.

Opportunity: Even with capability and intent, you need the materials, the working knowledge to go from a design on a screen to a stable, transmissible, weaponised agent, and the freedom to do it without being caught. That gap between "the AI wrote a genome" and "a functioning weapon" is still very wide. The Stanford work made phages that infect bacteria, not people. The distance from there to a human pathogen is not a weekend project.

So, should we be fearful? Here's my honest read.

The probability of a garage-built superbug next year is low. The trajectory of capability is the thing to watch.

We are removing the hardest barrier, which was always expertise, and we are removing it faster than the safeguards are being built.

That is the real story. Not the doomsday scenario. But, the slope.

There's a book doing the rounds right now that captures exactly this fear: Annie Jacobsen's Biological War: A Scenario.

Same author as Nuclear War: A Scenario, and it does the same thing here: a ticking-clock, hour-by-hour account of what happens after a biological agent is released.

Infrastructure buckling in days. Government struggling to keep functioning. Society sliding toward breakdown.

It reads like fiction, but it's built on interviews with the people who actually plan for this.

It makes the same point from the other side: the systems meant to catch this are thinner than you'd hope.

What this means for you

You're not going to build biosecurity policy from this newsletter.

But if you make decisions about:

1. Risk
2. Supply chains
3. People
4. Capital

Three things are worth holding onto.

Watch capability curves, not just incidents. By the time it's an incident, you're behind.

The signal is how fast the barrier is dropping.

Separate possible from probable, every time. Someone will always sell you the scariest version.

Your job is to ask what it would actually take, and who would actually want to.

And respect what an outbreak does to a fragile system.

The DRC is showing us, in real time, the difference between a threat on paper and a threat on the ground.

No vaccine, no trust, armed men at the roadblock. That's the environment where any pathogen, natural or designed, does its worst damage.

When was the last time your risk assessment told you the difference between what is possible and what is probable?

Because those two words, confused, are how smart organisations end up either panicking or asleep.

The barrier that kept this stuff hard was never the equipment. It was the knowledge. And knowledge is exactly what these tools hand out for free.

Ahmed
CEO Grey Dynamics
​Where headlines end, ground truth begins

​
​

Geopolitics Read More

Technology Read More