For the first time, humans design 16 entirely new viruses using artificial intelligence—a breakthrough with both promise and peril.
(Based on MSN original text and integrated authoritative reports from the same source)
Core Guide
This breakthrough in AI-designed viruses marks a major milestone in biotechnology, opening new pathways for disease research and drug development. However, the study also raises concerns: future AI systems could potentially engineer novel biological weapons, create lethal toxins or highly pathogenic viruses, and pose a threat to global security.
The AI-generated virus created in this study poses no threat to humans. It is fundamentally a bacteriophage—a naturally occurring virus that infects only bacteria. According to the latest report by The New York Times, researchers have for the first time used artificial intelligence to design an entirely new virus from scratch. While this breakthrough could enable scientists to develop novel approaches for disease research and infection treatment, it also raises serious biosecurity concerns.
1. Full Research Lifecycle
This project was a joint effort by Stanford University and the Arc Institute. The research team trained an AI system to learn and decode the genetic patterns embedded in DNA—the core molecule that carries all biological genetic instructions.
Once artificial intelligence masters genetic patterns, it can autonomously design and generate the complete genetic code for entirely new viruses. Researchers synthesize AI-generated virtual gene sequences into real DNA and insert them into bacteria for cultivation. The bacteria eventually produce novel viruses never before seen in nature—artificial viruses with full physiological capabilities to infect other bacteria and replicate independently.
The research team conducted experiments using two genomic large language models, Evo 1 and Evo 2, operating on principles similar to text-based LLMs like ChatGPT: while conventional models learn human language, the Evo series learns the "language of life"—DNA base sequences. The team trained these models on millions of natural phage gene sequences, enabling AI to autonomously identify genetic structural patterns and generate novel viral genomes.
AI generated hundreds of thousands of candidate viral genes. Researchers selected 302 promising sequences for laboratory synthesis and live testing, resulting in 16 bacteriophages that successfully survived and stably replicated—an experimental success rate of approximately 5.6%. The genomic sequences and structures of these three AI-designed bacteriophages do not overlap with any naturally occurring bacteriophages on Earth. In some cases, the engineered phages proved more effective at killing drug-resistant E. coli than their natural counterparts.
Project lead Brian Hoyer stated, "This study successfully completed a critical proof of concept, demonstrating for the first time that generative AI can design a complete, biologically active viral genome from scratch." The full paper was published in the top-tier journal *Science* on 2026/8/6.
Bacteriophage schematic diagram
II. Great Hope Brought by This Technology (Medical Value)
Conquering Superbugs
Millions die annually worldwide from drug-resistant bacterial infections caused by antibiotic failure. Traditional antibiotics are nearly ineffective against superbugs, making phage therapy a core alternative. While scientists previously spent years trial-and-erroring to find and engineer natural phages, AI can now rapidly batch-customize targeted bactericidal phages.
This experiment demonstrates that a cocktail of multiple AI-designed phages can rapidly inhibit E. coli strains resistant to natural phages, significantly reduce the likelihood of further bacterial mutation and resistance development, and offer a novel therapeutic option for difficult-to-treat drug-resistant infections.
Expanding the boundaries of biotech R&D
Experts in synthetic biology state that this technology overcomes natural evolutionary limits: many medically valuable phages cannot naturally merge due to geographic and evolutionary barriers. AI can directly integrate advantageous genes from different viruses to create custom biological carriers tailored for medical needs.
In the long term, this technology can be extended to develop biopharmaceuticals such as enzymes for correcting genetic disorders, antibodies for cancer immunotherapy, and targeted gene delivery vectors. It offers new solutions for various complex diseases and lays the technical foundation for designing complex living organisms through synthetic whole-genome engineering.
3. Deep Fears in the Global Scientific Community (Biosecurity Risks)
Misuse of technology to create biological weapons
A concurrent expert commentary in Science highlights the core risks: Generative AI now possesses the capability to design complete viral genomes, yet global biosafety oversight and ethical frameworks remain underdeveloped. While current research teams are only designing harmless phages, malicious actors or extremist groups with access to similar technology could modify training data to autonomously engineer novel pathogens capable of infecting humans with high lethality, thereby creating new biological weapons.
Thomas Ingelsby and Moritz Hank of the Johns Hopkins Center for Health Security issued an urgent warning: The question is no longer whether AI can design dangerous viruses, but how to prevent this technology from being used to cause mass harm. Existing regulations and laboratory screening protocols struggle to detect unknown high-risk gene sequences generated by AI in advance.
Lowering technical barriers and spreading risk
This research paves the way to lower the technical barriers for artificial virus creation. Historically, synthesizing entirely new functional viruses required top-tier laboratories and years of accumulated expertise. With the future widespread adoption of lightweight AI tools, small and medium-sized institutions—and even individuals—will have the opportunity to independently design and synthesize novel viruses, causing biosecurity control challenges to rise exponentially.
IV. Security Measures Implemented by the Research Team
To mitigate short-term risks, the research team has established multiple safety barriers:
The AI training database excludes all viral gene sequences capable of infecting humans, animals, or plants, as well as complex eukaryotic cells. Consequently, the model cannot generate viral genomes that cause human disease.
The experiment used a non-pathogenic standard strain of E. coli as the culture carrier throughout.
This experiment focused exclusively on bacteriophages with simple structures that infect only bacteria; we did not attempt to design complex zoonotic viruses.
However, experts emphasize that these safeguards only constrain this specific experiment and cannot eliminate risks at the source. As long as new training data is used, the same AI model can be redirected to design dangerous pathogens.
5. Industry Summary and Controversies
Proponents argue that AI-powered artificial phages are a breakthrough in combating the antibiotic crisis, with the potential to save millions of lives globally from drug-resistant infections.
Cautious scholars warn that this technology is a sharp double-edged sword: on one side lies a beacon of hope for saving lives in healthcare; on the other, an unlocked Pandora's box of biological risks. Nations must urgently establish unified global governance rules for bio-AI, strengthen gene synthesis screening mechanisms, and balance technological advancement with the safety of all humanity.