Antibiotic resistance
Bacteria are evolving resistance faster than new antibiotics arrive.
Open in the interactive tree →Every use of an antibiotic selects for resistant bacteria, and resistance genes spread between species. Few companies invest in new antibiotics because they are used sparingly and earn little. Without effective antibiotics, routine surgery, childbirth, chemotherapy and transplants become dangerous.
As of October 2026
WHO's October 2025 surveillance report found that one in six laboratory-confirmed bacterial infections worldwide was resistant to antibiotic treatment in 2023, rising to one in three in South-East Asia and the Eastern Mediterranean; over 40 percent of E. coli and over 55 percent of K. pneumoniae isolates resisted third-generation cephalosporins. The Lancet's GRAM study estimated 1.14 million deaths directly attributable to resistant bacteria in 2021 and forecast 39.1 million cumulative attributable deaths from 2025 to 2050. New antibiotic classes are rare, but the FDA approved two first-in-class oral antibiotics in 2025, gepotidacin (urinary tract infection in March, gonorrhea in December) and zoliflodacin (gonorrhea, December), and AI-guided discovery and phage therapy are advancing.
What is missing
- New antibiotic classes against Gram-negative bacteria, whose outer membrane keeps most drugs out
- A business model that pays for new antibiotics without rewarding overuse, for example subscription payments
- Fast, cheap diagnostics that tell a doctor within an hour which drug will work
- Surveillance data: 48 percent of countries reported no data in 2023
- Alternatives and prevention: phages, vaccines and stewardship in farming
Becomes possible once solved
- Safe surgery and chemotherapy in the long term
- New drug classes from AI design
- Phage therapy tailored to each infection
Open steps
- Passing the Gram-negative barrier High AI leverageDesign molecules that cross the outer membrane of Gram-negative bacteria and escape efflux pumps, where most candidates fail.
- New antibiotic classes High AI leverageFind new structural classes, not variants of old ones, against priority pathogens such as MRSA and carbapenem-resistant bacteria.
- One-hour susceptibility tests Medium AI leverageTell a doctor within an hour which drug will work, from genomes, spectra or growth signals.
- Forecasting resistance spread Medium AI leveragePredict which resistance genes will spread, from genomes and sparse surveillance data (48% of countries reported none in 2023).
- Designing phage therapy Medium AI leveragePick or design phages that overcome resistant bacteria and stay effective as the bacteria adapt.
Where AI could help
Medium AI leverage. AI has found new antibiotic candidates that work in mice, but payment models, Gram-negative delivery and clinical trials are bigger hurdles than discovery.
- Generative design of molecules that cross the Gram-negative outer membrane
- Predicting resistance spread from genomes and surveillance data
- Diagnostics that infer which drug will work from sequence or spectrum data
- Ranking drug and phage combinations against resistant infections
Shown so far
- In December 2023 Nature reported an explainable deep-learning screen of about 12 million compounds that found a new structural class active against MRSA and vancomycin-resistant enterococci, which cut infection in mice. source
- On 14 August 2025 Cell published MIT work in which generative AI evaluated over 36 million designed compounds and produced two lead antibiotics (NG1, DN1) that cleared drug-resistant gonorrhea and MRSA in mice. source
Prerequisites
- Evolution by natural selection1859
- Antibiotics1928
- Human microbiome2007-2012Gut-microbe studies show how antibiotics disturb the body's microbes and select for resistance
- AI protein design2023
Unlocks
- Infections under control2040s?