AI can now imagine antibiotics we never could

For more than a century, antibiotics have been one of our great triumphs—and a growing liability. These drugs are now faltering as the bacteria they were designed to defeat learn to fight back. Each year, millions of infections grow harder to treat and hundreds of thousands of people die because the pills and injections we rely on no longer work. It’s a crisis of our own making: overprescription, agricultural overuse, and decades of underinvestment have brought us here.

New hope that we can overcome this crisis has sprung from a lab at MIT, where a research team led by bioengineer James Collins recently reported in Cell that they have used generative artificial intelligence not simply to search for new antibiotics, but to design them from scratch. The team’s AI system imagined, molecule by molecule, new chemical structures that could pierce the defenses of two notorious foes: drug-resistant Neisseria gonorrhoeae and methicillin-resistant Staphylococcus aureus, better known as MRSA.

One compound, NG1, dismantled gonorrhea infections in both lab cultures and mouse models, apparently by targeting a bacterial protein essential for building the organism’s outer membrane. Another, DN1, shredded MRSA’s protective walls through a brute-force disruption of its cellular structure. Neither compound had existed before the AI conjured it.

Antibiotics are becoming the new designer product

Antibiotics have long been discovered by accident. Penicillin emerged from Alexander Fleming’s famously moldy petri dish in 1928. Streptomycin, the first effective tuberculosis drug, was unearthed in soil bacteria in the 1940s. For decades, researchers scoured nature for microbial compounds that happened to kill other microbes. But that well has run dry. In recent years, pharmaceutical companies have largely abandoned antibiotic research, unwilling to invest billions in drugs that patients take for only a few days and that lose effectiveness as resistance spreads.

AI is trying to break this impasse. Previous efforts, including the 2020 discovery of halicin and the 2023 unveiling of abaucin, used deep learning to comb through existing chemical libraries for overlooked gems. MIT’s latest work goes further: it turns AI into a designer. Instead of rummaging through known molecules, the system creates new ones from scratch by exploring vast new chemical spaces unmapped by humans. What once required decades of trial-and-error can now unfold in months with a fine-tuned system powered by GenAI.

Increasing resistance threatens routine treatment

Drug-resistant infections already kill more than1.2 million people worldwide each year, according to a 2019 Lancet study, and the toll is projected to climb dramatically by mid-century. The World Health Organization has warned that without new treatments, routine surgeries and chemotherapy could become perilous, as the risk of untreatable infection outweighs the benefits.

Gonorrhea is a case study in this unraveling. Once easily subdued with penicillin, the bacterium has evolved resistance to nearly every drug deployed against it. “We are currently down to one last recommended and effective class of antibiotics, cephalosporins, to treat this common infection,” warns the CDC. MRSA, too, haunts hospitals and communities alike, lurking on skin and in wounds, ready to seize any lapse in hygiene. For pathogens like these, the old arsenal is nearly empty.

Different routes to a whole new world

MIT is not alone in reimagining antibiotic discovery. At the University of Pennsylvania, researchers led by César de la Fuente have turned to evolutionary archaeology, mining ancient DNA from Neanderthals and woolly mammoths to extract genetic blueprints for antibacterial peptides. Others are building hybrid platforms that blend AI design with structural biology, accelerating the process of filtering millions of possible molecules down to a handful worth testing.

Global collaborations, too, are straining to fill the pipeline. Initiatives like CARB-X provide public-private funding to shepherd early-stage antibiotic candidates toward the clinic, but progress is slow and precarious. The economics remain brutal: developing antibiotics is expensive, but their sales are deliberately restricted to prevent resistance, leaving companies little incentive to stay in the game.

New hope in a dead field

It is tempting to frame MIT’s breakthrough as the dawn of a new golden age. British headlines have already done so, heralding AI-designed drugs as the revival of a field left for dead. Liam Shaw, in his book Dangerous Miracle, argues that antibiotics were always both blessing and curse: a miraculous reprieve that we squandered through hubris and neglect. For Shaw, the danger now is not just microbial evolution but human complacency—our assumption that science will always conjure another cure in time.

What makes MIT’s work striking is that it represents not another stroke of luck but a deliberate act of invention. For the first time, scientists are not merely discovering what evolution has hidden in the soil but authoring entirely new blueprints for life-and-death molecules.

A long way to go from here

Still, the leap from mouse models to human patients is enormous. NG1 and DN1 may never prove safe or effective in people. They could falter in clinical trials, prove too toxic, or encounter unforeseen resistance. Nonprofit ventures like Phare Bio, co-founded by Collins, are working to move them through the preclinical gauntlet, but years of testing lie ahead.

And even if they succeed, the broader question remains: how do we build a sustainable system for antibiotic innovation? Without new funding models (“pull” incentives like market entry rewards or government purchasing guarantees), scientific breakthroughs may wither on the vine.

Shifting to a higher gear

Perhaps what matters most is not the individual compounds but the proof of concept. Generative AI has shown it can expand the frontier of chemistry, inventing molecules no human had ever considered. If antibiotics mark the first battleground, the same approach could apply to antivirals, antifungals, and even entirely new classes of therapeutics.

The story of antibiotics has always been a story of time: bacteria evolve faster than we innovate. What MIT’s work hints at is a way to close that gap, to respond not in decades but in years or even months to the shifting microbial landscape.

That possibility is exhilarating, because it suggests that the era of serendipity may be giving way to an era of deliberate design. It is also sobering, because if history has taught us anything, it is that scientific breakthroughs are rarely permanent victories. Bacteria adapts, so the question is whether we, now armed with algorithms, can adapt faster.