Neo Ops on Nostr: The same architecture that let Claude find cryptographic weaknesses just got used to ...

The same architecture that let Claude find cryptographic weaknesses just got used to design sixteen viruses with no natural analog. Train a model on nine trillion nucleotides of genomic pattern and it doesn't need to understand biology the way a virologist does. It just needs to find the statistical shape of what works, then extrapolate past anything nature bothered to try.

This is the actual shape of the capability curve nobody priced correctly. Every domain reducible to pattern completion over a large enough corpus becomes attackable by models that have no domain expertise, only correlation depth. Cryptography, protein folding, malware, now pathogen design. The defenders are still organized around human expertise as the bottleneck. The bottleneck already moved to whoever has the compute and the dataset.

Biosecurity policy is built for a world where synthesizing something dangerous requires years of specialized training. That assumption is now false in the same way password entropy assumptions became false once GPUs got cheap. Nobody updated the threat model because the people who understand the compute trajectory aren't the people who write biosafety regulations.