NTI | bio Advances International Dialogue on AI Biosecurity Through Expert Working Groups

Biosecurity risks continue to evolve as artificial intelligence (AI) converges with other technological advancements in the life sciences. Designing effective risk mitigation solutions requires engagement from diverse experts from around the world. For this reason and in support of the AIxBio Global Forum, NTI | bio convened two technical working groups this fall to address critical challenges at the intersection of AI and the life sciences, bringing together more than 50 international experts across the two sessions from major AI companies, academic institutions, and biosecurity organizations to develop concrete solutions for safeguarding emerging technologies.

The Working Group on Guardrails for Biodesign Tools met virtually on September 25, followed by the Working Group on Horizon Scanning, Risk Assessment, and Evaluations on October 17. These meetings represent a significant step forward in NTI’s ongoing efforts to build international consensus and develop practical approaches for reducing risks associated with AI-bio capabilities.

The September meeting on biodesign tools focused on developing practical safeguards and managed access approaches for AI-enabled tools that can engineer biological systems. Key outcomes and recommendations included:

• Capturing metadata from biodesign tools: A project is underway to develop standards for capturing metadata, including what types of metadata can be captured and in what format. This metadata could assist DNA providers and others in determining potential harms by providing additional information about the user’s intentions in using the tool.
• Screening of inputs and outputs of biodesign tools: In addition to screening based on similarity to known pathogen or toxin sequences, future work will be needed to determine potential harms from protein structure or predicted function.

The October session on horizon scanning, risk assessment, and evaluations addressed the broader landscape of AI and biosecurity risks. Participants identified several priority areas:

• Horizon Scanning: In this rapidly changing field, it will be important to develop robust approaches to identify and assess technical advances in both AI and the life sciences that might change the risk landscape.
• Evaluation Frameworks: The group emphasized the need for robust, standardized approaches to evaluating AI models’ capabilities and risks, with particular focus on developing and sharing third-party evaluation mechanisms.
• Risk Assessment: The working group stressed the importance of creating and communicating a high-level vision of the biosecurity risks presented by AI advancements.

One discussant noted:

“A unique strength of this working group is a lot of expertise in areas that are adjacent to AI — synthesis screening, laboratory automation, involvement from people in industry — and I also think there’s some arguments that suggest that risks from AI will be much greater once we start to see a lot more developments from some of these other areas as well.”

NTI will convene regular meetings of both working groups through 2025, with the next full AIxBio Global Forum meeting scheduled for December 2024. The working groups will serve as vital venues to lay a high-level strategic approach to help focus the expert community as well as provide a place for experts to share ongoing projects and best practices.

These meetings build upon NTI’s work to safeguard AIxBio capabilities and broader work to strengthen biosecurity measures and reduce biological threats. The NTI report on “Developing Guardrails for AI Biodesign Tools” can be found here.