AI Hits Bottleneck in Virus Creation
A RAND experiment testing frontier LLMs on virus creation tasks found that AI boosted 4 out of 5 steps in the process. However, the models hit a wall at the molecular cloning stage, illustrating a key 'o-ring' limitation where a single bottleneck in a complex workflow can neutralize gains from AI elsewhere.
The January 2024 RAND Corporation report involved red-team exercises where researchers, role-playing as malign actors, attempted to plan biological attacks. The study found no statistically significant difference in the viability of plans generated with Large Language Model (LLM) assistance compared to those using only the internet, concluding that current AI does not meaningfully increase the operational risk. The 'o-ring' concept originates from economist and Nobel laureate Michael Kremer's theory on economic development, named after the component failure that caused the Space Shuttle Challenger disaster. The theory posits that in complex production chains, every step must be performed proficiently; a single failure can render the entire effort worthless, magnifying the impact of the weakest link. This principle resonates in cell and gene therapy (CGT) manufacturing, where key bottlenecks include process scalability and the high cost of goods, often driven by manual labor and a lack of robust automation. While AI can optimize parts of the workflow, these physical and logistical constraints act as the limiting 'o-rings' that hinder overall efficiency and accessibility of advanced therapies. A primary bottleneck for AI's application in bioprocess optimization is not algorithmic limitation but data infrastructure. The lack of standardized
