AI Is Designing New Antibiotics to Fight the Superbugs We Can't

Part 2 of the AI Receipts series.

Drug-resistant infections are one of the scariest slow-moving threats in medicine. Bacteria evolve faster than we invent new drugs to kill them, and the antibiotic discovery pipeline has been running dry for decades. It is expensive, it is slow, and the economics have pushed most of the industry out of the field.

In 2020, a team at MIT pointed a machine-learning model at a library of more than 100 million chemical compounds and surfaced a brand new antibiotic they named halicin. In lab testing, it killed bacteria that shrug off our existing drugs, including several dangerous, treatment-resistant species.

In 2023, an improved model uncovered an entirely new structural class of antibiotics that had been hiding in plain sight inside a vast chemical library.

Then, in 2025, the work leveled up again. Instead of searching existing libraries, the team started designing new antibiotic molecules from scratch with generative AI, building compounds atom by atom. Two of the lead molecules killed drug-resistant gonorrhea and MRSA, two pathogens the world urgently needs answers for.

Notice what did not change through any of this. Humans still run the experiments. Humans still verify every result in the lab. What the AI does is expand the search and design space those humans can work in, from thousands of candidate molecules to hundreds of millions, and now to molecules that never existed before.

When someone tells you AI has no real-world value, this is the kind of thing they are waving away. Not a chatbot writing marketing copy. New weapons against infections that are, right now, killing people our current drugs cannot save.

/Andy

Sources

Next
Next

The AI That Solved a 50-Year-Old Problem and Won a Nobel Prize.