Alexander Fleming — discoverer of penicillin in 1928 — warned in his 1945 Nobel acceptance lecture: "It is not difficult to make microbes resistant to penicillin in the laboratory by exposing them to concentrations not sufficient to kill them, and the same thing has occasionally happened in the body … There is the danger that the ignorant man may easily underdose himself and by exposing his microbes to non-lethal quantities of the drug make them resistant." He was exactly right. By the late 1940s, penicillin-resistant Staphylococcus aureus was clinically established. By the 1960s, methicillin-resistant S. aureus (MRSA). By the 2010s, carbapenem-resistant Enterobacteriaceae and pan-drug-resistant Acinetobacter baumannii. The pipeline of new antibiotics had simultaneously collapsed (fewer than half a dozen genuinely novel classes have entered clinical use since 1970). Antibiotic resistance is the slow-motion public-health emergency of the 21st century, and it is natural selection in real time, in a Petri dish or a hospital ward.
Antibiotics work by attacking targets that bacterial cells need and human cells either don't have or use differently. β-lactams like penicillin disrupt cell-wall synthesis; aminoglycosides, macrolides, and tetracyclines bind to bacterial ribosomes and disrupt translation; fluoroquinolones inhibit DNA gyrase; sulfonamides block folate synthesis; glycopeptides like vancomycin attack cell-wall precursors. Each target buys you a window of clinical efficacy, and each window is closed by the same evolutionary logic. Bacteria acquire resistance through some combination of enzymatic inactivation of the drug, modification or replacement of the drug's target, efflux pumps that expel the antibiotic faster than it can enter, and reduced membrane permeability. What turns the local arms race into a global one is horizontal gene transfer: resistance genes ride on plasmids, transposons, and bacteriophages that move between bacterial species and even between genera, so a resistance allele that arises in an environmental microbe can reach a clinical pathogen and become a global epidemic strain within years rather than the geological times that govern vertical inheritance.
The selection pressure driving this is enormous and largely unforced. Humans use roughly two hundred thousand tonnes of antibiotics per year, and the majority of that goes into food animals, often at sub-therapeutic doses ideal for selecting resistance rather than killing pathogens. Hospitals concentrate vulnerable patients, antibiotic use, and resistance genes in close quarters and have become hotspots for resistant-organism evolution; wastewater, soil, and agricultural runoff carry resistance genes back into the environmental reservoirs that started the cycle. The global burden, in the Lancet's 2022 estimate, was around 1.27 million deaths directly attributable to antibiotic-resistant bacterial infections in 2019 — comparable to HIV or malaria — with several million more in which resistance contributed. The structural problem is that the antibiotic R&D pipeline has been collapsing for forty years: short treatment courses produce low margins, the resistance pressure that follows market entry erodes long-term value, and several major pharmaceutical companies have exited antimicrobial development entirely. The threat Fleming flagged in his Nobel lecture is materializing exactly as he warned, and the new pipeline of phage therapies, AI-discovered compounds like halicin (Stokes et al., 2020), and bacterial vaccines is racing a curve that has been bending the wrong way since the 1980s.
Antimicrobial resistance is now treated as a top-tier global-security threat, not merely a clinical nuisance. The WHO maintains a priority-pathogens list to steer research, and the influential 2016 O'Neill review projected, on current trends, up to ten million deaths a year by 2050 — a contested but agenda-setting figure. The policy response runs on two tracks: stewardship — prescribing more narrowly and for shorter courses, cutting agricultural use, deploying rapid diagnostics so the right drug is given the first time — and incentives to refill the pipeline, including subscription-style payments that reward a new antibiotic for being available rather than for sales volume. Machine learning has begun to surface genuinely novel candidates from a chemical space too vast to screen by hand. Resistance can't be defeated, only managed — the goal is to bend the curve, not break it.