Every Cure, a Philadelphia-based nonprofit, runs an AI system every month that scores roughly 4,000 existing, already-approved drugs against more than 18,000 known diseases — about 75 million possible drug-disease combinations — hunting for cases where a medicine developed for one condition could be repurposed to treat a completely different one that currently has no good treatment. Three years ago, generating that kind of shortlist for a single disease took the organization’s researchers about 100 days of manual literature review. Its AI pipeline now does it in about 17 hours, according to reporting from CNBC earlier this year that has continued to shape how the drug-repurposing field talks about AI’s near-term potential.
Drug repurposing itself isn’t new — sildenafil became Viagra after failing as a heart medication, and thalidomide, notorious for causing birth defects when used in pregnancy, was later repurposed as an effective treatment for multiple myeloma and leprosy-related complications. What’s new is using generative AI language models to systematically scan the entire universe of approved drugs against the entire universe of known diseases, rather than relying on individual researchers stumbling onto promising connections through clinical observation or serendipity.
Why Repurposing Is Cheaper Than Inventing
Developing a brand-new drug from scratch typically takes over a decade and costs, by most industry estimates, well over a billion dollars once failed candidates are factored in — and the majority of experimental drugs that enter human trials never reach approval. Repurposing an already-approved drug skips most of that risk: its safety profile, dosing, and manufacturing are already established, meaning a repurposed drug can potentially reach patients through a much shorter and cheaper regulatory path, provided a repurposing sponsor can generate credible evidence of efficacy for the new indication.
That economic logic is especially important for rare diseases, which individually affect too few patients to justify the enormous cost of dedicated drug development but collectively affect an estimated 30 million Americans. Every Cure’s stated goal is to enable treatment for 15 to 25 diseases through repurposed drugs by 2030 — a modest-sounding number set deliberately conservatively against an enormous space of possible matches.
Federal Money Is Backing the Bet
The approach has attracted serious federal interest: Every Cure is set to receive $48.3 million from ARPA-H, the Advanced Research Projects Agency for Health, to build out its AI-driven drug repurposing platform further — a substantial vote of confidence from a federal agency explicitly designed to fund high-risk, high-reward biomedical research bets. The organization has also entered a collaboration with Predictive Oncology to specifically pursue repurposing opportunities for cancer patients, extending its approach into oncology.
Where the AI Actually Helps, and Where It Doesn’t
Every Cure has been notably explicit that its platform is not intended to operate without human oversight — the organization describes its approach as “human-in-the-loop,” where AI generates ranked candidate matches and flags promising signals, but its scientific staff still evaluate the biological plausibility of each match, design the studies needed to test it, and navigate the regulatory and clinical trial process to actually get a repurposed drug into patients’ hands. That distinction matters because the hardest part of drug repurposing has historically not been generating hypotheses — researchers have long had informal lists of promising repurposing candidates — but rather funding and running the clinical trials needed to prove a repurposed use actually works, something an AI matching algorithm alone cannot do.
The Skeptical Counterpoint
Health policy researchers who study drug repurposing caution that a faster hypothesis-generation pipeline doesn’t automatically translate into more approved treatments, because the bottleneck for repurposing has typically been financial rather than intellectual: pharmaceutical companies have limited commercial incentive to fund trials for already-generic drugs that can’t support the pricing power of a new patented medicine. Every Cure’s nonprofit structure is explicitly designed to route around that incentive problem, but critics note the model still depends on continued philanthropic and federal funding rather than a self-sustaining commercial engine, raising questions about how many of its AI-flagged candidates can ultimately be funded through to clinical proof and regulatory use.
What’s Next
With ARPA-H funding now committed and a growing set of partnerships including its oncology collaboration with Predictive Oncology, Every Cure’s near-term test is converting its AI-generated shortlist of drug-disease matches into actual clinical evidence, one disease at a time. The organization’s leadership has framed 2026 through 2030 as the period in which its model needs to prove out its first handful of real repurposed treatments — the point at which the promise of a 17-hour AI search either translates into patients actually being treated, or remains an impressively fast way to generate hypotheses that still take years to validate.
