Newly created viruses are a warning. We still have a window to stop AI-enabled bioweapons

Bulletin of the Atomic Scientists | Steph Guerra

An August 6 study in the journal Science demonstrated that artificial intelligence models can design novel viruses as effectively as nature, raising fears of future AI-enabled bioweapons. While the experiment only created bacteria-infecting pathogens, the rapid advancement of these digital tools could soon allow malicious actors to design highly transmissible, human-infecting viral agents capable of sparking global pandemics.

Historically, modern biological attacks have relied on non-transmissible agents like anthrax or ricin. To counter this emerging threat, biosecurity experts advocate for physical chokepoints like mandatory gene synthesis screening for DNA and RNA orders, alongside strict digital controls on curated biological datasets. These defenses are not foolproof. Consequently, establishing pathogen early warning systems and publicly demonstrating rapid response capabilities, such as the Coalition for Epidemic Preparedness Innovations' fast-tracked vaccine initiatives, are critical for deterrence. Ultimately, preventing global proliferation requires sustained political leadership and international coordination, particularly between the United States and China, before AI capabilities advance further.

Comment

The integration of generative artificial intelligence into biological design fundamentally challenges the verification architectures established under the 1972 Biological Weapons Convention (BWC). Traditional non-proliferation frameworks rely on monitoring physical precursors, specialised laboratory equipment, and known pathogenic strains. However, AI-driven de novo synthesis allows actors to design functional, novel pathogens from digital sequence data, bypassing established watchlists entirely. This shift from physical material acquisition to digital design generation renders traditional export control lists, such as those maintained by the Australia Group, increasingly obsolete.

Consequently, national intelligence agencies will face severe attribution challenges, as the distinction between natural zoonotic spillovers and engineered outbreaks becomes nearly indistinguishable. This diagnostic ambiguity will likely force defence planning to pivot from threat-specific countermeasures to platform-based, broad-spectrum medical defences. Ultimately, the speed of AI-driven mutation design will compress the decision cycle for deploying rapid-response countermeasures like those coordinated by the Coalition for Epidemic Preparedness Innovations (CEPI).

Strategic Question for Discussion

If AI-driven de novo synthesis allows adversaries to bypass the physical precursor controls of the Australia Group, what alternative multilateral mechanisms can realistically prevent the digital distribution of lethal genetic blueprints?

The trajectory indicates a shift toward mandatory cryptographic watermarking and access-control protocols embedded directly within commercial DNA synthesis hardware. Furthermore, establishing international standards for secure, closed-loop cloud environments for biological foundation models offers a potential pathway to restricting digital blueprints without choking legitimate scientific research. This transition effectively moves the regulatory boundary from physical borders to the digital compute layer.

Share your assessment in the comments below.

To read in details click here