A safety-centric perspective on innovation and risk in the use of artificial intelligence in genomics.

Generative genomic AI models pose significant safety risks, including data quality issues, biases, and dual-use concerns, especially in critical fields like medicine and pathogen engineering.

• Why it matters: Addressing these risks is crucial as the accessibility of generative models increases, potentially leading to misuse and harm in sensitive applications, highlighting a gap in current safety frameworks.
• What they did: The article examines how misuse risks develop throughout the innovation pipeline, focusing on dual-use governance and infrastructure hardening, and offers a toolkit for stakeholders to balance innovation with safety.
• The result: This work enables scientists, regulators, and policymakers to foster beneficial genomic AI innovations while implementing robust safeguards to prevent misuse and ensure safety.