Strengthening nucleic acid biosecurity screening against generative protein design tools
- Bruce Wittmann ,
- Tessa Alexanian ,
- Craig Bartling ,
- Jacob Beal ,
- Adam Clore ,
- James Diggans ,
- Kevin Flyangolts ,
- Bryan Gemler ,
- Tom Mitchell ,
- Steven T. Murphy ,
- Nicole E. Wheeler ,
- Eric Horvitz
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.
What does it mean to paraphrase a protein?
This explainer video unpacks the science behind paraphrasing a protein—a defining technique of Microsoft’s Paraphrase Project. Researchers discovered it is possible to use advanced open-source AI tools to reformulate toxic proteins, potentially preserving their function while disguising their form. The goal? To test whether existing DNA screening systems could detect these “paraphrased” threats. The results were concerning: thousands of synthetic variants slipped through undetected. In response, a cross-sector team developed and deployed updated algorithms, patching a critical gap and bolstering detection capabilities. By acknowledging the dual-use nature of powerful tools, this work offers foundational guidance and emerging best practices for safer, responsible innovation in synthetic biology.
Learn more about the Paraphrase Project
Paper: Strengthening nucleic acid biosecurity screening against generative protein design tools | October 2, 2025 | Science
Feature: The Paraphrase Project: Designing defense for an era of synthetic biology | October 6, 2025
Podcast: Ideas: More AI-resilient biosecurity with the Paraphrase Project | October 6, 2025 | Eric Horvitz, Bruce Wittmann, Tessa Alexanian, and James Diggans
Blog: When AI Meets Biology: Promise, Risk, and Responsibility | October 6, 2025 | Eric Horvitz