AI models generate viable bacteriophages in lab test, raising dual-use concerns

Researchers reported that two AI models were tasked with generating complete genomes for viable bacteriophages—viruses that infect and destroy bacteria. The models used the ΦX174 bacteriophage, known for its ability to infect and destroy E. coli, as a reference.

The models produced approximately 700,000 potential genome designs, from which researchers selected 285 candidates they deemed most promising based on computational criteria.

The team synthesized DNA molecules corresponding to the selected designs and inserted them into E. coli bacteria. Within a short period, 16 of the Petri dishes showed clear signs of viral activity as the engineered bacteriophages infected and replicated within the bacteria.

Among the viable viruses produced, some demonstrated greater effectiveness at attacking E. coli than the original ΦX174 strain, indicating the AI-generated designs were not only functional but in some cases improved upon the natural template.

The researchers characterized the work as both promising for applications such as phage therapy and concerning due to the potential for misuse in biological engineering or biowarfare scenarios.