You’ve probably never thought about it, but the most dangerous thing on the internet right now might not be a data leak or a ransomware attack. It’s a sequence of letters — A, T, C, G — that can be pasted into a text file. Because AI just learned how to write those letters in a way that creates a functional virus.
In September 2025, researchers showed that generative AI models could design viral genomes that actually work. The same technology that powers ChatGPT can now design a pathogen. Think about that.
The tool that could cure the next superbug is also the tool that could start the next pandemic. And the code is already out there.
Most people think biosecurity is about locked labs, hazmat suits, and armed guards. That’s 20th-century thinking. The 21st-century threat is digital. A viral genome is just data. And data can be copied, hidden, and run on a laptop in a coffee shop.
A bioinformatician told me, ‘We spent billions on BSL-4 labs. We forgot to lock the laptop.’ That’s the uncomfortable truth.
The most dangerous pathogen is not a virus in a vial. It’s a sequence in a text file.
The irony is that the same AI models that can design a deadly virus can also design a phage that kills antibiotic-resistant bacteria. The same science, the same code, the same math. The only difference is intent.
This means the debate about open-source AI isn’t just about copyright or job displacement. It’s about whether we hand the keys to a biological weapon to anyone with an internet connection. The risk is not hypothetical. The code may already be circulating.
Neutrality is death. The same AI that can save lives can take them. We have to choose which side we’re on — and we have to choose now.
The next pandemic won’t start with a leak from a lab. It’ll start with a commit to a public repository. The question is: are we ready for that world?
FAQ
Q: Isn't this fear-mongering? AI models are not that accurate yet.
A: No. The 2025 demonstration showed functional viral genomes. Accuracy is improving rapidly. The barrier is not technical capability but ethical oversight. The fear is justified because the capability is real.
Q: So what should we do? Regulate open-source AI models?
A: Not necessarily regulate open-source, but we need new norms around model release, data access, and screening of biological sequences. The same way we screen for nuclear secrets, we need to screen for dangerous DNA sequences. But it's a global challenge.
Q: The contrarian take: Actually, this is a good thing because it democratizes biological research.
A: Democratization has a dark side. While it enables small labs to design therapies, it also enables bad actors. The potential for good is enormous, but the risk of misuse is equally enormous. We need to balance openness with safeguards.