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

The Bulletin of the Atomic Scientists | Steph Guerra

Artificial intelligence models capable of designing novel viruses have demonstrated the capacity to match or exceed natural evolutionary pathways, raising severe biosecurity risks. This breakthrough, published in the journal Science on August 6, indicates that future viral bioweapons could become significantly easier to engineer and potentially trigger transmissible human pandemics.

To mitigate these emerging threats, policymakers are advocating for mandatory gene synthesis screening among scores of US companies alongside strict digital controls on laboratory-validated datasets. However, physical and digital barriers remain vulnerable to bypass. Advanced adversaries can still culture pathogens independently or deploy jailbroken models to circumvent safety filters, rendering traditional containment strategies insufficient. Consequently, establishing pathogen early warning systems and robust international coordination, particularly between the United States and China, is essential to build a resilient, layered global defense that deters attacks through rapid response capabilities like those funded by the Coalition for Epidemic Preparedness Innovations.

Comment

The transition from static pathogen databases to generative biological design models like the Evo genomic foundation model fundamentally alters the requirements for biosecurity verification. Traditional screening protocols, which rely on matching orders against known threat lists like the Australia Group common control list, cannot detect entirely novel, AI-generated genetic sequences. This capability gap exposes a critical vulnerability in commercial gene synthesis pipelines, where benchtop DNA synthesisers like the DNA Script SYNTAX system operate outside centralised cloud-monitoring frameworks. Consequently, the proliferation of these benchtop DNA printers bypasses the physical chokepoints managed by the International Gene Synthesis Consortium.

This evasion mechanism relies on machine learning algorithms optimising nucleotide sequences to obscure homologous matches to regulated agents like the Ebola virus while preserving the functional phenotype of the designed pathogen. Addressing this bypass involves transitioning International Gene Synthesis Consortium protocols to functional prediction models that evaluate the pathogenic potential of a sequence rather than its sequence identity. This transition relies on integrating structural biology tools like AlphaFold to predict protein-receptor interactions before the International Gene Synthesis Consortium authorises production.

Strategic Question for Discussion

If benchtop DNA synthesisers like the DNA Script SYNTAX system become widely distributed, how can the International Gene Synthesis Consortium enforce compliance without relying on centralised cloud-based kill switches?

The trajectory indicates that relying solely on cloud-based kill switches will fail as offline, open-source synthesis software proliferates. My assessment is that compliance verification will require hardware-level cryptographic signatures embedded directly into synthesis microfluidic chips. This approach would allow the International Gene Synthesis Consortium to verify the authorisation of genetic orders even on completely air-gapped systems.

Share your assessment in the comments below.

To read in details click here