New NTI Framework Tackles Misuse Risks in AIxBio Tools

Today, NTI | bio released a new paper, A Framework for Managed Access to Biological AI Tools. The paper calls on tool developers and other AIxBio stakeholders to implement a managed access approach—allowing only validated users to access those biological AI models that contribute to biosecurity risks—to help reduce the risk that such tools will be deliberately misused to cause harm.

The convergence of AI and the life sciences has driven the development of biological AI tools with a range of beneficial applications. However, some of these tools have also raised concerns among biosecurity experts that they could be misused, including by making it easier to engineer dangerous pathogens.

Informed by interviews with more than 20 experts in biosecurity, AIxBio tool development, and bioscience research, the framework adopts two central principles to maximize the potential benefits of AIxBio tools while reducing risks:

1. Access should be tiered based on the tool’s risk level.
2. The need for security should be balanced with the need to provide equitable access.

This new paper offers recommendations for funders, model developers, and platforms that host models to support implementation of managed access. Best practices for managed access should be developed and tailored to the needs of different AIxBio communities:

1. Funders should offer low- or no-cost access to infrastructure to developers who implement appropriate managed access; fund the development of new managed access platforms, where needed; and fund technical projects and workshops to support development of managed access tools and best practices.
2. Model Developers should use a tiered risk framework to assess their tools during development; implement managed access approaches that are appropriate to the risk level of their tools, and record lessons learned and work with other model developers to identify best practices.
3. AIxBio platforms and hosts should support innovation by enabling tools to be discovered, verified, used, adapted, compared with similar tools, and maintained over time; adopt transparent, defensible, and consistent criteria for users to establish legitimacy; and expand access through secure, user-friendly APIs.

“We know that biological AI model developers want to do the right thing, and to be responsible. If we can provide a resource for this growing community, that would benefit them by making it easier to safeguard their models,” said Jaime Yassif, PhD, a senior advisor to NTI | bio. “We can apply managed access in a targeted way to the most powerful models to guard against the most consequential risks. 

Managed access is essential to reducing biosecurity risks from misuse of AIxBio tools. Building on the framework proposed in the paper, NTI will work with other stakeholders to further develop clear, tiered risk frameworks for biological AI tools. NTI will also engage technical partners to explore the possibility of building a managed access platform that demonstrates these elements. As this framework becomes more widely used, funders, governments, and other stakeholders should create incentives to encourage broader adoption.