fox files

DARPA Uses AI to Push Viral Pandemic Outbreak Modeling From Weeks to Days

Speed is being prioritized over scrutiny, with AI-generated models designed to justify interventions before they can be meaningfully challenged.

The U.S. military is funding artificial intelligence (AI) systems designed to drastically accelerate viral outbreak modeling—compressing a process that typically takes weeks into something that can be produced in days, then used to steer real-world interventions.

In other words, the faster the model, the less time there is to question whether the response is justified at all.

This acceleration follows DARPA’s already-documented pre-COVID pandemic infrastructure designed to turn digital genetic sequences into synthesized viruses and mass-produced mRNA countermeasures on a fixed timeline.

---

DARPA’s Stated Problem: Pandemic Models Were Brittle, Opaque, & Slow

According to a December Science publication:

As SARS-CoV-2 radiated across the planet in 2020, epidemiologists scrambled to predict its spread—and its deadly consequences. Often, they turned to models that not only simulate viral transmission and hospitalization rates, but can also predict the effect of interventions: masks, vaccines, or travel bans.

But in addition to being computationally intensive, models in epidemiology and other disciplines can be black boxes: millions of lines of legacy code subject to finicky tunings by operators at research organizations scattered around the world. They don’t always provide clear guidance. “The models that are used are often kind of brittle and nonexplainable,” says Erica Briscoe, who was a program manager for the Automating Scientific Knowledge Extraction and Modeling (ASKEM) project at the Defense Advanced Research Projects Agency (DARPA).

The Defense Advanced Research Projects Agency’s (DARPA) own program manager is conceding that the models used to steer COVID-era responses were fragile and difficult to interpret.

Meaning: they’re not trying to slow down or restrain model-driven policy after COVID.

They’re trying to make the same kind of decision machinery run faster.

There’s “real potential” for them to speed up model building during an outbreak, says Mohsen Malekinejad, an epidemiologist at the University of California San Francisco who helped evaluate the ASKEM products. “In a pandemic, time is always our biggest constraint. We need to have the information. We need to have it fast,” he says. “We simply don’t have enough data-skilled modelers for every single emergence, or every different type of public health need.”

The Program: AI-Generated Outbreak Models on Demand

“Launched in 2022, the $29.4 million program aims to develop artificial intelligence (AI)-based tools that can make model building easier, faster, and more transparent.”

DARPA funded infrastructure that standardizes and accelerates outbreak modeling.

The emphasis is on speed, reproducibility, and usability by non-specialists, allowing policy-relevant models to be generated quickly, even when underlying assumptions are incomplete or contested.

How It Works: Papers & Notebooks → Equations → Models

“The program’s AI tools automate that coding, allowing researchers to construct, update, and combine models at a higher level of abstraction.”

By removing much of the technical friction involved in model construction, these tools make it easier to generate outbreak models that carry institutional weight, even when the scientific grounding is thin or uncertain.

“ASKEM teams designed AI systems that can consume scientific literature… and extract the equations and knowledge needed to create or update a given model.”

Scientific literature is converted directly into reusable model components, giving machine-parsed interpretations of research the ability to propagate quickly into decision-making frameworks.

“One ASKEM project developed a way to ingest those notebooks, extract the underlying mathematical descriptions, and turn them into a model.”

Informal reasoning and exploratory notebook work can be elevated into deployable models at speed, reducing the distance between preliminary thinking and authoritative outputs.

Intervention-Focused Modeling

“The resulting model integrated the viruses’ different transmission and seasonal patterns, while gauging the effects of interventions such as wearing masks and testing.”

The system is designed to evaluate intervention scenarios alongside disease dynamics, embedding policy considerations directly into the modeling process.

“Testers were asked to model the impact of a vaccination campaign on the cost of hospitalization for hepatitis A in a state’s unhoused drug users.”

These tools are oriented toward applied governance questions—cost, targeting, and campaign impact—rather than purely descriptive epidemiology.

The Speed Claim: 83% Faster

“In the final results, testers found that the ASKEM tools, when pitted against standard modeling workflows, could create models 83% faster.”

Model generation is fast enough to fit within political and media timelines, reducing the opportunity for external review before results are acted upon.

“They were able to build practically useful models in a 40-hour work week for multiple problems.”

Once speed ceases to be the limiting factor, the pressure shifts toward rapid implementation rather than careful validation.

‘Transparency’ as an Internal Confidence Signal

“Because of the ASKEM models’ enhanced transparency, testers also found that decision-makers would be more confident in ASKEM’s outputs than in those of traditional models.”

Here, “transparency” functions less as a safeguard and more as a confidence amplifier for officials.

By making models legible enough to satisfy internal review, the system reduces friction within institutions, allowing officials to act more quickly while unresolved uncertainties remain embedded in the outputs.

Intended Users: Health, Defense, & Intelligence Agencies

“DARPA is working to find agencies or programs within the health, defense, and intelligence communities that might want to take advantage of ASKEM.”

Outbreak modeling is being positioned as a permanent national-security capability, integrated alongside defense and intelligence functions rather than treated as an ad hoc public-health exercise.

Bottom Line

DARPA is building a system that converts literature, assumptions, and exploratory analysis into outbreak models fast enough to guide interventions in near real time.

When speed is treated as the primary constraint, the window for scrutiny, dissent, and meaningful challenge necessarily collapses before those models are used to justify action.