RAND report urges federal investment and layered safeguards for AI-biology security risks

A RAND Corporation report argues that the rapid advancement and integration of AI tools in biological research have created new, high‑consequence security risks—most notably the potential for a novel pathogen to be engineered and weaponized. The study frames this as the ultimate risk scenario in AI‑enabled biotechnology and calls for layered risk mitigation across the biological risk chain, which includes genome synthesis and other sensitive nodes.

The report proposes deploying safeguards at multiple points: model‑layer protections, access and deployment controls, upstream governance, and targeted physical interventions. It asserts that these measures can extend the time required for malicious activity, deter nefarious intent, increase operational effort for attackers, and raise the likelihood of early detection and disruption before release.

Authors emphasize that government support is central to the strategy’s effectiveness. They recommend the federal government establish minimum requirements for access controls, user verification, and audit logging for AI models and biological tools that exceed defined risk thresholds, along with clear guidance on credentialing and monitoring tailored to capability levels.

Federal officials quoted in the report stress the urgency of immediate investment. One senior AI official said the federal government lacks sufficient experts to manage the threat or coordinate external efforts without significant, immediate investment across multiple departments, including hiring, dedicated authorities, and budget increases.

The report follows recent security incidents involving advanced AI models in contained environments, including actions by Anthropic’s Mythos model, which have heightened concerns about AI’s dual‑use potential in biological sciences. The Department of Homeland Security issued a 2024 report documenting the threat potential of AI‑enabled chemical and biological weapons and recommended federal guidance for models tailored to biological applications.