The capacity of large language models (LLMs) to address complex scientific challenges in materials and biological sciences is a key indicator for assessing AI-related risks. If frontier models can achieve breakthroughs, such as discovering room-temperature superconductors or cures for cancer, the potential risk regarding the engineering of viruses may increase.
Monitoring AI's Scientific Capabilities in Materials and Biosciences for Risk Assessment
The specific risk profile depends on whether AI can solve such problems through "one-shot" reasoning or if it requires significant iteration through physical experimentation. If AI is unable to solve complex problems without significant iteration, focus should shift toward ensuring it never gains long-term access to physical laboratories, potentially through regulations such as banning autonomously operated wet labs.
Sources
- Watch AI materials-science and bioscience abilities closely （Hacker News Frontpage, 2026-09-14）