There is hardly an adult American who at some point hasn’t needed antibiotics — for bronchitis or a urinary tract infection, for example, or something more serious like pneumonia or blood poisoning. Seldom do we think about their long and storied history.
Nearly a century ago, a stroke of luck changed the course of medicine. In 1928, Scottish microbiologist Alexander Fleming returned from vacation to find that one of his bacterial culture plates had been contaminated with mold. Around the mold was a clear halo where the bacteria had died. The mold was Penicillium notatum, and the substance it produced — which Fleming named penicillin — launched the modern antibiotic era.
Since then, antibiotics have transformed once-deadly infections into treatable conditions and saved countless lives. However, evolution never sleeps, and as antibiotics were used — and overused — in medicine and agriculture, bacteria mutated and adapted. Today, antibiotic-resistant “superbugs” threaten to undo decades of progress. Some infections that were once easily cured now withstand multiple drugs. Meanwhile, the pipeline for new antibiotics has slowed to a trickle.
For years, experts have warned that we are losing ground. Now, artificial intelligence is flipping the script.
In October, a research team at MIT reported a breakthrough. It used generative AI to create entirely new antibiotics effective against two especially dangerous pathogens — Neisseria gonorrhoeae, which causes gonorrhea, and methicillin-resistant Staphylo
Antibiotic resistance arises when bacteria evolve ways to evade the drugs meant to kill them. A 2019 report estimated that drug-resistant infections already kill more than a million people annually; by 2050, that number could reach 10 million worldwide. Yet, new antibiotics are desperately scarce. Most “new” drugs are minor variations of old ones, which means bacteria often become resistant to them quickly.
Generative AI promises to break that cycle by designing molecules unlike anything seen before.
The MIT team built an AI framework capable of sifting through unimaginably vast chemical landscapes. Rather than searching existing chemical libraries — a limited universe — the system generates brand-new molecules, predicts their antibacterial potential, and filters out those that are toxic or unstable.
The researchers used two approaches. The first, a fragment-based strategy, began with a commercial database of billions of possible fragments, from which AI identified one promising starter structure, called F1, that showed strong activity against N. gonorrhoeae. From that single fragment, the algorithms generated seven million molecules. After several rounds of screening, researchers selected 80 for laboratory testing. Only two could actually be synthesized — a reminder that chemistry poses its own constraints — but one of them, dubbed NG1, proved remarkably potent. It killed drug-resistant gonorrhea not only in lab dishes but in infected mice.
More striking, NG1 appears to work in a completely new way. It disables a protein essential for building the bacterium’s outer membrane. That mechanism is distinct from those of existing antibiotics, which could make it harder for bacteria to outmaneuver.
The second approach took creativity to an extreme. Here, the AI was told only to generate chemically plausible molecules with antibacterial potential — no starting fragment, no blueprint. The result: 29 million hypothetical compounds. From these, the team tested 22 in the lab. Six showed strong activity against MRSA, and one — DN1 — successfully treated skin infections in mice.
Of the 24 compounds synthesized across both projects, seven killed bacteria without harming human cells, and NG1 and DN1 stood out as the most promising. A nonprofit partner, Phare Bio, is now working to refine them for possible human trials.
The road ahead is long. Many compounds with promise in the lab falter in clinical trials. It can take years — even decades — to turn a molecule into a marketable drug. And bacteria will continue to evolve, no matter how inventive our tools become.
However, for the first time in decades, the field has momentum. Generative AI can explore chemical spaces that human scientists could never hope to survey. It can identify unexpected molecular structures, propose drug candidates at lightning speed, and point researchers toward mechanisms that bacteria have not yet learned to defeat.
The antibiotic crisis remains one of the gravest challenges in modern medicine. Fleming’s serendipitous discovery of penicillin may soon be joined by breakthroughs driven not by chance but by algorithms trained to explore the unknown. AI could help usher in a new era of antibiotic discovery, restoring an edge humanity has been steadily losing.