New roadmap targets AI-enabled bioweapon risks with nine safeguards

The same artificial intelligence capabilities accelerating drug discovery could enable bad actors to design biological weapons — but a comprehensive safeguard framework could prevent catastrophic misuse, according to a new analysis.

A RAND Corporation report released this week proposes nine mitigation strategies spanning government oversight, scientific research controls, and public health preparedness. The framework addresses a growing concern: AI is rapidly expanding the pool of potential bioweapon developers beyond state actors to include individuals with limited expertise.

Why it matters

The biological research tools driving medical breakthroughs — from protein design to drug candidate discovery — operate on the same underlying technologies that could be weaponized. As AI agents become more capable of autonomous research, the barrier to entry for dangerous biological work continues to fall, creating an urgent need for coordinated safeguards that don't stifle legitimate science.

Layered defense across four fronts

The RAND approach targets four intervention points: restricting access to dangerous information, disrupting access to necessary materials, detecting misuse signals early, and deterring malicious activity before it starts.

"AI capabilities in biology are advancing really fast, but luckily most dangerous thresholds have not yet been crossed," RAND lead author Steph Guerra said. "We need this layered system of mutually reinforcing approaches."

The report frames public health preparedness not merely as outbreak response but as active deterrence — a shift in how health security intersects with AI governance.

The open science dilemma

A central tension emerges around biological datasets and research tools. While pharmaceutical companies maintain proprietary control over their AI systems, academic institutions and research organizations frequently publish open-source biological data and tools.

"Right now, data for biology is freely out there and open because biology [is] for discovery and saving lives. It's not warfare first, which is a good thing," Guerra explained.

The report proposes tiered access systems that preserve scientific collaboration while restricting the most misuse-prone datasets to verified researchers. Guerra advocates for "big moonshot projects and biodata factories" that generate disease-curing data within controlled access frameworks.

Industry acknowledgment grows

Major AI companies are beginning to address these risks publicly. Anthropic announced this week that Claude successfully designed protein binders against 14 targets — a significant drug development milestone — while simultaneously restricting access to these research biology capabilities in its most powerful model.

"Without robust safety measures, they could enable bad actors to perform dangerous research, such as the development of bioweapons," Anthropic stated, acknowledging the dual-use nature of increasingly autonomous AI research tools.

The RAND team has been specifically investigating how AI agents lower expertise requirements for biological work, making sophisticated capabilities accessible to users without traditional scientific training.

The report's recommendations extend beyond frontier AI companies to encompass the broader ecosystem of biological research, public health infrastructure, and government regulation — reflecting the reality that no single sector can address these risks in isolation.

These details were first reported by Axios.