Special Episode: Recipes for Future Plagues

Episode Summary

Executive Summary: The episode argues that USAID’s new Deep Vision program, while aimed at preventing pandemics, could inadvertently create a major biosecurity threat by publicly posting genomes of newly discovered, pandemic-capable viruses. Through a detailed discussion of gene drives, virus hunting, and genomic characterization, the hosts weigh public-health gains against the risk of empowering tens of thousands of people to reconstruct dangerous pathogens.

Main Topics: Deep Vision and the central biosecurity concern (Priority: 5/5): The conversation focuses on a new USAID pandemic-prevention program that would sequence, characterize, and publicly share data on unknown wildlife viruses. The concern is that identifying and publishing genomes of the most dangerous viruses could function like distributing blueprints for weapons of mass destruction. Gene drives, CRISPR, and defense-favoring biology (Priority: 4/5): Kevin Esvelt explains gene drives, CRISPR-based recursive genome editing, and why some technologies "favor defense" because misuse can be reversed or blocked. This establishes the framework for distinguishing between tools that can be safely countered and those that can’t. Historical precedent: resurrecting and sharing deadly virus genomes (Priority: 5/5): The discussion uses the 1918 influenza genome and smallpox as examples of how publishing pathogen sequences can increase proliferation risk. The hosts debate whether the scientific gains justify making these genomes available online. The mechanics of virus hunting and characterization (Priority: 5/5): Deep Vision’s process is broken into virus hunting, sequencing, and lab characterization. The characterization step—testing binding, infectivity, replication, and transmissibility—creates the greatest security risk because it identifies which viruses are most dangerous and worth weaponizing. Arguments for and against information hazards (Priority: 4/5): Sam and Rob present the concept of information hazards: the value of knowledge can be negative when it enables harmful actors. They compare viral genome release to nuclear attention hazards, where revealing capability can motivate proliferation and imitation. Policy responses and alternative pandemic preparedness (Priority: 4/5): The episode ends with proposed mitigations: redirecting funds away from characterization, strengthening surveillance, using emergency vaccine protocols, expanding mRNA vaccine capacity, and contacting USAID, WSU, and lawmakers to reconsider the program’s focus.

Key Arguments: Publishing genomes of newly discovered pandemic-capable viruses could dramatically lower the barrier for tens of thousands of labs to synthesize dangerous pathogens from scratch. The 1918 influenza genome was published even though the virus could be reconstructed, and the number of people able to do so has grown substantially as DNA synthesis has become cheaper. Gene drives are not the central problem because they spread slowly, are visible, and can be countered with immunizing reversal drives; unlike pathogens, they favor defense. Deep Vision’s virus-hunting and sequencing alone could support surveillance and broad-spectrum vaccine research, but the characterization phase is what creates a proliferation risk. Knowing the exact genomes of dangerous zoonotic viruses provides little vaccine-development advantage in practice because modern vaccines can be designed quickly, while testing, approval, and distribution remain the bottlenecks. A malicious actor could exploit public genome data to make multiple pandemic-capable viruses and release them simultaneously across travel hubs, making deliberate pandemics worse than natural spillovers. The existence of an open, authoritative government-backed list of dangerous viruses makes coercive threats more credible, increasing informational and psychological harm beyond direct use. Pandemic-prevention policy should be treated more like nuclear nonproliferation: limit access, classify high-risk experiments, and discourage global dissemination of enabling blueprints.

Data Points: Deep Vision budget share: less than 0.1% of USAID budget - Used to emphasize that the program is small but potentially high-risk. PREDICT viruses identified: about 1,200 previously unidentified viruses - Referenced as the predecessor program’s output. Deep Vision target output: 10,000 to 12,000 viruses over 5 years - Rob argues Deep Vision aims to scale virus discovery far beyond PREDICT. DNA synthesis cost decline: factor of 1,000 decrease - Explains why reconstructing viruses from genomes has become much easier. Potential reconstructors today: 20,000 to 50,000 people - Estimate of how many could synthesize a virus from a published genome now. 1918 flu death estimate: up to one in 30 living humans - Used to underscore the severity of the resurrected virus. 1918 flu plausible death toll if released today: 200,000 to 10 million - Kevin’s rough confidence interval given modern mitigations. Historic influenza assembly protocol: 1999 - First detailed protocol for assembling influenza viruses. First virus synthesized from synthetic DNA: 2002 - Poliovirus is cited as an early proof of feasibility. Smallpox vaccine stockpile: 350 million doses - Used to argue the world has substantial defenses against variola. H5N1 case fatality rate mentioned: 60% - Discussed as a frightening example of engineered transmissibility risk. COVID vaccine development time: 342-ish days - Used to argue that vaccine design is fast but testing, approval, and distribution are slow. Omicron spread benchmark: 102 days since sequencing - Used to argue that 100-day vaccine goals may still be too slow. Nuclear-armed states: 9 commonly acknowledged nation-states - Compared against the number of actors who might gain access to pandemic blueprints. Estimated credible actors for a released dangerous virus: roughly a thousandfold increase - Sam’s interpretation of how public genomes expand the threat surface.

Pivotal Quotes: "the genomes of previously unknown pandemic-grade viruses that we have no defenses against." — Sam Harris: Describing the alleged Deep Vision risk of posting virus blueprints online. "Anything that is slow, obvious, and easily blocked isn't much of a weapon." — Kevin Esvelt: Explaining why gene drives can be comparatively defensible and reversible. "The value of information can be negative." — Kevin Esvelt: Core concept behind treating viral genome publication as an information hazard.

Implications: The episode frames pandemic-genome disclosure as a potential proliferation event, not just a research gain. If listeners accept the argument, they may support tighter controls, selective secrecy, and a shift toward surveillance and vaccine readiness over public release of enabling data.