Stanford University announced on August 6th a breakthrough in biotechnology. Their research team introduced a new bacteriophage, capable of precisely eradicating Escherichia coli (E. coli), synthesized in the laboratory using the “Evo 2” generative artificial intelligence (AI) model. However, this innovation aimed at tackling the global superbug crisis has stirred up strong reactions in the international scientific community and biosecurity field.
In the face of the escalating issue of antibiotic resistance, the team led by Assistant Professor Brian Hie and graduate student Samuel King from the Chemical Engineering Department at Stanford University targeted bacteriophages, viruses specialized in infecting and killing bacteria.
Using the generative bio AI model Evo 2, the research team employed the classic bacteriophage ΦX174 with a genome length of about 5,400 base pairs as a blueprint to generate new bacteriophage gene sequences. Researchers evaluated and selected candidate sequences through computational frameworks, followed by chemical synthesis and laboratory testing.
The team ultimately synthesized nearly 300 AI-designed bacteriophage variants and, through in vivo testing, identified 16 bacteriophages that effectively inhibit E. coli, with some variants showing higher adaptability in experiments than the natural ΦX174. The experiments further confirmed that combining these AI-generated bacteriophages into a “cocktail therapy” could significantly reduce the probability of E. coli developing antibiotic resistance, offering a new direction for exploring phage therapy for antibiotic-resistant infections.
Despite the enormous medical potential, Stanford’s decision to fully open-source the code and model weights of Evo 2 has been highly questioned by several authoritative experts.
Pioneering synthetic biologist J. Craig Venter expressed extreme concern, pointing out that when generative AI possesses the ability to randomly write genes of living viruses, the most terrifying risk lies in its application to Gain-of-Function research. If malicious individuals were to apply this technology to smallpox, anthrax, or influenza viruses, they could potentially create deadly pathogens completely unfamiliar to the human immune system and lacking ready-made vaccines and drugs.
Cybersecurity and biodefense experts highlight that under the open-source framework, any malicious actor with basic equipment could use Adversarial Fine-tuning techniques to feed human pathogenic virus data back into the model. Security defenses originally set by authorities could be easily bypassed within a very short period of time.
Regarding the controversy crossing the boundaries of medical innovation and biosecurity, experts Thomas Inglesby and Moritz Hanke from the Center for Health Security at Johns Hopkins University bluntly stated in the top international journal “Science” that the current global regulatory system is seriously lagging behind.
Experts point out that the existing biosecurity review mechanisms are mostly based on a “known dangerous pathogen list,” but tools like Evo 2, a generative genomics tool, create “entirely new sequences never before seen in nature,” for which current regulations are nearly non-existent.