A common flu variant tweaked to increase lethality. A synthetic coronavirus programmed to evade current vaccines. A strain of pneumonic plague modified to be resistant to antibiotics. As fears grow that AI has lowered the barriers to engineering pathogens, some experts have been calling for a Manhattan Project-scale effort to counter the potential biological risk. In recent months, philanthropists, entrepreneurs, and government funders—judging the best defense against AI-enabled pandemics to be AI itself—have invested hundreds of millions of dollars in high-tech projects to develop rapid, scalable responses to such scenarios.
Yet, most medical countermeasures cannot protect people if they aren’t willing to accept them, and while AI can accelerate their development, it cannot manufacture the trust needed for their successful deployment. The current momentum in biodefense risks overindexing on technical solutions to a human problem.
If an AI-enabled vaccine became available within days of a future outbreak, the question would remain: Will people get it?
Supply-side preparedness. Vaccination campaigns during outbreaks in past decades have sometimes faltered, and factors such as rare adverse events have driven public resistance. Even technological success in the face of a real health crisis is no guarantee of public trust. Yet, recent investments have prioritized these supply-side solutions to biological threats, with large AI companies directing extraordinary capital and compute toward AI-powered preparedness initiatives.
In May, OpenAI launched a dedicated biodefense program that includes investment in the development of AI-designed antibodies against novel pathogens, while Biohub, joined by Meta, Google DeepMind, and federal agencies announced in October a $1.8 billion public-private partnership with the eventual goal of conducting computer-based biological experiments. These efforts reflect a broader push by tech companies and leading scientists to compress the time between the beginning of an outbreak and when governments can mount an effective response.
The new initiatives have been launching at a rapid clip.
The World Health Organization (WHO) announced a target in October to detect novel pathogens within 24 hours of emergence by 2030. Other organizations are developing computer modeling to rapidly assess pathogenicity and virulence digitally. Last year, for example, the company Valthos, said it received $30 million in a funding round led by the OpenAI Startup Fund to build “AI systems to rapidly characterize biological sequences and update medicines in real time.” And multiple startups are proposing to use AI to design vaccines and therapeutics to deploy as soon as “almost instantly.”
If these solutions work—and hopefully they do—they could yield candidate countermeasures to a newly discovered threat before it spread between more than a handful of people.
While this scenario is far from today’s reality, discussions have already begun on designing proper governance and oversight for this future. However, even as biosecurity experts have focused on issues such as how to manage access to AI-response technology or how to equitably distribute any vaccines that result from these rapid efforts, they have rarely addressed public acceptance.
The scientific evidence needed to establish that a countermeasure works may be very different from the evidence the public needs to believe it is safe and necessary. An AI-modeled vaccine may offer what the public perceives as an untested cure to an insignificant threat.
That’s happened before—even when an outbreak has already caused notable harm.
The early waves of COVID devastated communities and the rapid introduction of a safe and effective mRNA vaccine was a marvel of biotechnology. Still, the rollout was met with fierce resistance and it eventually plateaued with around 70 percent of the US population receiving the complete primary vaccine series, below the estimate of upward of 85 percent that Anthony Fauci, a key presidential advisor during the pandemic, suggested would be needed to achieve herd immunity. Since then, demand for updated COVID vaccine formulations has collapsed, driven by a broader vaccine fatigue that is now also manifesting as hesitation in routine childhood immunizations.
Vaccines designed to intercept biological threats before they cause widespread harm—the type of rapid response the new crop of biodefense funders hope to enable—may face even greater resistance. In survey research asking Americans if they would accept a Centers for Disease Control and Prevention (CDC)-recommended vaccine for bird flu, a disease that has so far only occasionally infected humans, less than four in 10 said they would.
Why trust in new vaccines can drop. At least three findings suggest that public acceptance of new AI-enabled countermeasures should not be an afterthought. First, novelty carries a penalty. A recent experiment presented adults with a fictitious influenza outbreak and a rapidly developed vaccine. Confidence in both safety and effectiveness was significantly lower for a novel platform than a traditional one. Introducing a trusted messenger did not close the confidence gap. A novel AI-designed product or platform will face this penalty immediately.
Second, confidence rarely returns once lost. Over the fall, OpenAI introduced 722 manuscripts detailing progress on some of the hardest problems in mathematics. The day after the announcement, the company withdrew three of them. For a company using its technology to solve the near-impossible, their success rate was extraordinary. But public health does not get the luxury of retractions. With vaccine and therapeutic introductions, even minor mishaps can cause persistent or permanent declines in confidence, especially when audiences are primed for skepticism. In 2013, the Japanese government suspended their proactive recommendation of the HPV vaccine to investigate unconfirmed media reports of adverse events among recipients. Vaccination rates initially plummeted from about 70 percent to less than 1 percent. Despite over a decade of extensive scientific investigations confirming the vaccine’s safety and the government’s eventual reinstatement of the recommendation, vaccination rates have yet to fully recover.
Third, acceptance can suffer when speed outpaces evidence of harm. In 1976, the federal government mounted a swift and massive vaccination campaign following a swine flu outbreak at Fort Dix, a US Army base in New Jersey. When the anticipated spread of disease never materialized, rare vaccine side effects became the dominant narrative. Routine influenza vaccination campaigns suffered in the following years, with patients frequently citing the fiasco to justify refusal.
Similarly, measles is now returning to communities that had forgotten its severe impacts, highlighting how vaccine uptake has declined as evidence of a pathogen’s harm has become less salient.
During COVID, public health authorities had months of accumulating disease burden, public discussion, and clinical evidence before introducing a novel prophylactic countermeasure. AI-driven solutions will usher in an exciting new era of health progress, but they are designed to disrupt this already accelerated timeline, potentially causing evidentiary and technical milestones that far eclipse social ones.
Closing these gaps does not require slowing development. Instead, those deploying the capital currently flooding into AI biodefense must expand their definition of success beyond speed and experimental endpoints. This is particularly relevant for the philanthropic sector driving much of this mobilization. Philanthropic goals are already aimed at measurable impact, but translating technical moonshots into real-world outcomes requires social infrastructure.
The good news is frameworks to do this already exist in adjacent fields, with researchers actively developing strategies in behavioral science, community health, and other disciplines. Pre-bunking, for example, works like a cognitive vaccine, introducing individuals to defensive strategies against misinformation prior to exposure to inoculate them against future deception. Similarly, the trusted-messenger model recommends bypassing top-down expert lectures about the merits of vaccination and instead leveraging respected local figures and peers to tackle the underlying causes of hesitancy. Still, changing minds is notoriously difficult, and how contemporary vaccination attitudes will respond to AI countermeasures co-developed with Silicon Valley remains an open question.
Funders seeking to maximize their technical breakthroughs must ensure their capital also flows to this less glamorous work of social preparedness.
Trust in public health sits at historical lows, and trust in AI is lower still. We cannot assume that technical virtue alone will override this deficit. Securing public confidence requires early structural investment in the institutional credibility of the biological AI sector—long before a novel pathogen emerges.
Leave a Reply