AI-designed novel viruses
BiosecurityComments
This struggle with stability is exactly how we developed synthetic vaccines for rare zoonotic diseases. Refining the assembly process usually leads to safer, more stable delivery vehicles for medicine.
The claim that creation always precedes control is a bit tidy. With CRISPR, we had the control mechanisms essentially built into the system from the start; the problem was the lack of ethical consensus, not technical control.
But look at the recent protein folding breakthroughs... the speed of these AI models means we are skipping the trial-and-error phase entirely. It makes the catch-up period the OP mentioned much more dangerous since the jump is so sudden...
This isn't just a biosecurity issue. It is the death of the biological moat. If anyone with a high-end GPU can design a pathogen, the concept of a secure lab becomes a quaint relic of the twentieth century.
The paper actually specifies these are attenuated viruses designed for targeted gene delivery. The biosecurity risk stems from the underlying generative model being open-sourced, not the specific viruses created in this study.
To what extent does the AI handle the actual folding and stability of the capsid? Creating a sequence is one thing; getting it to assemble into a viable virion in vitro is where these designs usually fail.