Scott Fay, VP, Informatics
Earlier this year I had the opportunity to participate in a panel discussion about AI and biosecurity at the annual SynBioBeta meeting. Those of us closest to this issue, including my colleagues and I at Ansa, feel a great sense of urgency about improving biosecurity practices for DNA synthesis. The prominence of this topic at the conference was a good reminder of how the synbio community can pull together to help drive needed, meaningful change.
The rise of generative AI models is creating new opportunities for bad actors to design harmful biological agents and circumvent nucleic acid sequence screening tools. These models present sequence screening software developers (such as Aclid, RTX, and IBBIS) with serious new technical challenges. However, this community of committed engineers, scientists, and policymakers is working hard to keep pace; for example, see this recent Frontiers in Bioengineering and Biotechnology special issue on biosecurity. Addressing the increased risk that AI design tools present to biosecurity through better sequence screening tools is necessary, but not enough. We also need to address how to induce nucleic acid synthesis providers to actually do biosecurity screening and create a better system for verifying legitimacy of end users.
In the DNA synthesis field, biosecurity is largely built on the honor system. While the possession of certain nucleic acids in the U.S. is regulated by the Federal Select Agent Program and international shipment is regulated by Commerce Control Lists, the scope of these lists is very limited, and defined by taxa, not biological function. Moreover, DNA synthesis providers are not regulated in how they screen sequences and users of those nucleic acids; screening policy only comes in the form of non-binding guidance with no mechanism for enforcement. For providers, aside from our own collective interest in not enabling a serious biological attack, there is little other incentive to comply with the guidance. Disincentives include lost revenue and increased operational costs. We need a better way to induce all synthetic DNA providers to conduct rigorous screening for every single order. There is legislation designed to do just this working its way through the U.S. Senate (S.3741) and California Legislature (AB1864). A strong inducement needs to be put in place, and we support efforts like these to establish binding, enforceable screening requirements across the industry.
Once we identify a sequence of concern in screening, we then have to evaluate whether the customer has a legitimate use for it. The botulinum toxin gene, for example, encodes for one of the most potent known toxins, but is also important in legitimate pharmaceutical research. We can’t simply refuse to make any sequence of concern, or we’d hold back progress in biotechnology, medicine and fundamental biology research. How does one establish the legitimacy of a customer? Current recommendations involve checking contact information like address, organization, and publication history, but the process isn’t codified in any way. At Ansa, when we discover a sequence of concern, we reach out to the customer to establish their legitimacy – consistent with current industry practice, since no codified standard yet exists. This gap underscores the need for a clear, standardized, industry-wide framework. Organizations like Sentinel Bio are working to fund efforts to improve governance around customer screening and verification processes.
I am grateful that the team at Ansa is supportive of building out a robust biosecurity process. Biosecurity screening can too easily be seen as a barrier to ease of ordering and therefore a barrier to revenue. It only takes one vendor prioritizing expediency over security to let something dangerous slip through. That’s why those of us who feel a personal responsibility for biosecurity are hungry for much stronger governance. We want to serve our customers and support great science without putting society at risk.