AI Designs Novel Viruses, Sparking Biosecurity Alarms and Medical Hope

AI's Leap into De Novo Viral Design

In a groundbreaking achievement, researchers at Stanford University and the Arc Institute have leveraged artificial intelligence to design 16 entirely new, functional viruses that do not exist in nature. This marks the first instance where AI has been successfully used to create complete viral genomes from scratch. The AI models, named Evo1 and Evo2, function similarly to large language models like ChatGPT, but instead of processing text, they were trained on vast datasets of genetic sequences from millions of organisms.

The scientists focused on designing bacteriophages, which are viruses that specifically infect bacteria and pose no threat to humans, animals, or plants. From thousands of AI-generated genome combinations, approximately 300 were selected for laboratory synthesis and testing, with 16 proving to be viable viruses. These AI-designed bacteriophages demonstrated the ability to infect and kill E. coli bacteria, even overcoming resistance in some strains. This capability suggests a significant potential for developing new therapies to combat antibiotic-resistant superbugs.

The Dual-Use Dilemma: Promise and Peril

While the creation of AI-designed viruses holds immense promise for medical advancements, particularly in addressing the growing crisis of antibiotic resistance, it also raises profound biosecurity and biosafety questions. Experts like Professor Tom Inglesby and Dr. Moritz Hanke from the Johns Hopkins Center for Health Security emphasize that the ability to compose viral genomes using generative AI now exists, but the necessary governance to safely steer this technology does not. The concern is that AI, if trained on genetic data of dangerous pathogens, could be used to design more harmful viruses.

The rapid pace of AI's progress in biology is outpacing current regulatory frameworks. The convergence of AI and synthetic biology enables the design of "sequences of concern" that may not appear on existing lists of regulated pathogens but could encode functions that increase virulence, transmissibility, or resistance to treatment. Moreover, AI chatbots could lower the barrier to entry for malicious actors by providing technical guidance for developing biological weapons.

Navigating the Uncharted Waters of AI Biosecurity

The potential for AI to design entirely novel DNA sequences, undetectable by current screening tools, presents a significant challenge to existing biosecurity measures. Current international legal obligations, such as the Biological Weapons Convention, were established decades before modern synthetic biology and AI advancements. This highlights a critical need for modernized regulations and robust screening mechanisms for synthetic DNA manufacturers.

To mitigate these risks, several measures are being considered:

• Expanding screening to detect novel viruses designed by AI.
• Developing techniques for identifying "warning shots" or failed attempts at engineering pandemics.
• Restricting access to AIs with hazardous biological knowledge to vetted researchers.
• Limiting the open publication of research into techniques for evaluating the potential for viruses to cause harm in humans, especially regarding transmissibility.
• Applying rigorous risk-benefit analysis to viral gain-of-function research.

The developers of the Evo models intentionally excluded genetic code for viruses that infect plants, humans, or animals from the AI's training to reduce the risk of designing dangerous viruses. However, the broader implications of AI's growing capabilities in biological design necessitate a proactive and comprehensive approach to global biosecurity.