Artificial intelligence-driven protein design has fundamentally changed what is possible in protein engineering. Deep learning models can now generate entirely novel sequences that fold into defined structures, enabling advances in therapeutics, vaccine development, and industrial biotechnology. For biosecurity specifically, designed proteins offer new opportunities: capturing and detecting biological agents, developing novel binders against viral surface proteins, and accelerating pandemic preparedness. Yet the same capabilities introduce new risks. AI-generated proteins may be functionally equivalent to known toxins while sharing little sequence similarity, rendering current homology-based screening blind to such designs. The wide availability of open-source tools further lowers the barrier to misuse. Mitigation requires layered strategies that together can deter misuse without stifling innovation. Here, we review the current landscape of generative protein design, assess its dual-use implications, and discuss proportionate mitigation strategies that balance open scientific progress with biosecurity.
Brackmann et al. (2026) studied this question.