#463 — Privatizing the Apocalypse

Episode Summary

Executive Summary: Sam Harris and Rob Reed discuss how Reed’s biosecurity advocacy began with fiction and research, then centered on the canceled U.S. Deep Vision program—a proposed virus-hunting and characterization effort Reed argues could have massively increased pandemic and bioterror risk. The conversation covers how public and political pressure helped defang and eventually kill the program, and why Reed sees future AI-enabled biology as making such dangers more widespread.

Main Topics: How Rob Reed became focused on biosecurity (Priority: 4/5): Reed traces his interest in bio-risk to writing a sci-fi novel, then interviewing scientists, starting a podcast, and giving a TED talk; this evolved into sustained advocacy on pandemic and bioterror risk. What Deep Vision was designed to do (Priority: 5/5): Deep Vision combined virus hunting, characterization experiments, and public release of viral genomes—an approach Reed argues could have amplified risk rather than reduced it. Why virus hunting can be dangerous (Priority: 5/5): Reed argues that sampling remote viruses and moving them into laboratories creates leak risk, since all labs leak at some rate and remote reservoirs may be safer than human-staffed facilities. The role of public pressure in stopping the program (Priority: 4/5): A network of activists, experts, lawmakers, and officials quietly pushed back, leading first to the program being defanged and later formally canceled. The broader biosecurity landscape (Priority: 4/5): The discussion compares Deep Vision to other programs and actors, including Wuhan Institute of Virology and prior U.S.-funded virus discovery efforts, while stressing how few actors can mount projects of this scale. AI and future proliferation of dangerous biology (Priority: 4/5): Reed warns that AI will lower barriers to sophisticated biological work, increasing the number of people capable of creating or misusing highly dangerous pathogens.

Key Arguments: Deep Vision was not merely surveillance; its design increased the chance of accidental release by moving unknown viruses from nature into leak-prone labs. Characterizing newly discovered viruses can be misleadingly reassuring or dangerously amplifying, because it can create famous pathogens studied widely without delivering real vaccine efficacy or safety data. Publishing viral genomes and a list of the most dangerous candidates could have distributed powerful offensive capabilities to thousands of people worldwide who might be able to recreate them. The project’s worst-case consequences could have rivaled or exceeded COVID-19, especially if multiple engineered or discovered pathogens emerged simultaneously from different locations. A very small number of well-funded, well-connected people can unintentionally create civilization-scale risk even when acting with benevolent intent. The cancellation of Deep Vision is evidence that public scrutiny and bipartisan pressure can sometimes prevent dangerous biosecurity policies. AI is likely to multiply the number of people able to reason about, design, or weaponize biology, making governance and oversight more urgent.

Data Points: Deep Vision budget: $125 million over 5 years - The proposed USAID program’s funding level. Planned virus discovery: ~10,000 undiscovered viruses - Estimated number of viruses Deep Vision aimed to find in remote field sites. Countries targeted: 12 developing countries - Reed says the program planned work in 5 African, 5 Asian, and 2 Latin American countries. Regional breakdown: 5 Africa / 5 Asia / 2 Latin America - Geographic distribution of planned Deep Vision fieldwork. People able to create viruses: ~30,000 - Kevin Esvelt’s estimate of people with tools and know-how to conjure viruses using reverse genetics/viral rescue. Public response timeline: Days before posting the episode, Russia invaded Ukraine - The invasion distracted attention and delayed visible response to the warning campaign. Program status: Effectively defanged, then formally killed in Sept. 2023 - Reed says pressure first halted work and later led to official termination. Testimony timing: Summer 2022 - Rand Paul’s hearing where Kevin Esvelt testified and discussed Deep Vision. Virus discovery scale under prior program: ~1,200 novel mammalian viruses - Reed references USAID’s earlier PREDICT program as a comparison.

Pivotal Quotes: "In a worse case, [it had] the potential to cancel civilization." — Rob Reed: Describing how a biosecurity expert characterized Deep Vision’s worst-case risk. "You were potentially giving the killing power of a nuclear arsenal to 30,000 completely unvetted strangers throughout the world." — Rob Reed: On the danger of publishing viral genomes and enabling reconstruction of dangerous viruses. "The problem was more or less passed." — Rob Reed: Summarizing the point when the project had effectively been defanged by pressure.

Implications: The episode argues that biosecurity failures can arise from well-intentioned public-health programs, so oversight should prioritize worst-case misuse and lab-leak risk. As AI lowers barriers to advanced biology, these concerns become more urgent for governments, funders, and researchers.