Additional Coverage:
I recently had the chance to speak with Eric Nguyen, co-founder and CEO of Radical Numerics, a pioneer in teaching artificial intelligence to write DNA. Nguyen envisions AI revolutionizing medicine by crafting personalized treatments and targeting diseases like cancer through precise DNA editing. Yet, during our conversation, he raised a chilling possibility that sounds more like a plot from a sci-fi thriller than straightforward biology: the emergence of “deepfake viruses.”
Unlike digital deepfakes that target phones or computers, a deepfake biological virus strikes at the very cells within our bodies. What makes it so dangerous is its ability to disguise itself-rearranging its genetic code just enough to evade detection by our immune system and existing screening technologies, all while retaining its harmful functions.
“AI can design DNA sequences that act like viruses but scramble their genetic letters so they don’t match anything we’ve seen before,” Nguyen explained. This stealthy rearrangement means that both natural immune defenses and artificial detection tools might fail to recognize the threat.
This concept, once sounding like a distant worry, has now entered reality. On August 6, researchers from Stanford and the Arc Institute unveiled a groundbreaking study published in Science.
They used generative AI models-specifically Evo 1 and Evo 2, developed by Nguyen’s team-to design entirely new viral genomes from scratch. These novel sequences were synthesized in the lab, resulting in 16 functional bacteriophages capable of infecting and killing E. coli bacteria.
Importantly, these are not human-infecting viruses but bacteriophages, viruses that target bacteria. Given the global crisis of antibiotic-resistant infections claiming millions of lives annually, this advancement holds promise for developing new phage therapies. The researchers took precautions by excluding sequences from viruses that infect humans, animals, plants, or fungi, and conducted their work under strict biosecurity measures.
Still, the implications are profound. The study proves that AI can generate fully functional viral genomes from mere data prompts.
While the current focus is on bacteriophages, the same technology could be used to create other viruses-if misapplied. This raises serious biosecurity concerns, especially since not all research teams worldwide operate under strict ethical guidelines.
Currently, safeguards exist: DNA synthesis companies screen orders to prevent the creation of dangerous genetic material. But Nguyen warns that AI’s ability to design viruses with rearranged genetic codes can potentially bypass these filters, creating a “biological deepfake” that slips through unnoticed.
This emerging threat challenges existing biosecurity frameworks. While the U.S. has restricted federally funded gain-of-function research, these rules don’t cover AI-generated novel viruses, nor do they apply globally. As biosecurity experts note, the capability to craft viral genomes with AI is here, but effective governance and oversight have not yet caught up.
On the bright side, AI-driven DNA design offers enormous potential benefits-from early disease detection and faster personalized drug development to rapid antiviral production during pandemics. Nguyen himself has shifted toward more cautious openness; while his earlier models were publicly released, newer, more powerful ones are now held back due to safety concerns.
The bottom line: our world’s defense systems were built assuming virus creation is difficult. AI is rapidly changing that assumption, making it easier to design new viruses-both good and bad. The harmless bacteriophages in this study serve as proof of concept, but also as a stark reminder that the tools enabling this breakthrough could be exploited with far more dangerous consequences.
With “Pandora’s box” now open, the imperative is clear: we must develop equally swift and sophisticated defenses, including AI-driven medicine design, to counter emerging threats before they spread. As Nguyen emphasizes, building these protective capabilities is not just prudent-it’s essential.
It looks like we’ll need them sooner than we hoped.