Project Summary
Advances in artificial intelligence (AI)–assisted protein engineering are enabling breakthroughs in the life sciences but also introduce new biosecurity challenges. Synthesis of nucleic acids is a choke point in AI-assisted protein engineering pipelines. Thus, an important focus for efforts to enhance biosecurity given AI-enabled capabilities is bolstering methods used by nucleic acid synthesis providers to screen orders. We evaluated the ability of open-source AI-powered protein design software to create variants of proteins of concern that could evade detection by the biosecurity screening tools used by nucleic acid synthesis providers, identifying a vulnerability where AI-redesigned sequences could not be detected reliably by current tools. In response, we developed and deployed patches, greatly improving detection rates of synthetic homologs more likely to retain wild type–like function. One way in which companies and governments can limit potentially harmful biological research is by restricting the commercial synthesis of DNA that encodes particular proteins or is from select pathogens. However, screening methods are not necessarily designed to detect engineered sequences. Wittmann et al. worked with four commercial DNA synthesis companies to stress test and develop patches for screening methods to greatly improve their ability to identify sequences that should be restricted. Their results demonstrate that proteins of concern can be flagged by updated software even if their sequences have been altered using protein design methods.
Access the paper
Wittmann, B. J., Alexanian, T., Bartling, C., Beal, J., Clore, A., Diggans, J., Flyangolts, K., Gemler, B. T., Mitchell, T., Murphy, S. T., Wheeler, N. E., and Horvitz, E. Strengthening nucleic acid biosecurity screening against generative protein design tools. Science 390, 82-87 (2025). DOI: 10.1126/science.adu8578 (bibtex)
Supplementary Materials
Framing Study
Wittmann, B. J. and Horvitz, E. (2023). Adversarial AI-Assisted Protein Design for Evading Hazard Detection: Methods, Results, and Mitigations. Technical Report, Office of the Chief Scientific Officer, Microsoft, Redmond, WA, October 27, 2023. (bibtex)