Every few decades, medicine faces a reckoning. In the 1940s, Alexander Fleming's mouldy petri dish gave the world penicillin and rewrote the rules of surgery, childbirth, and warfare. For the better part of a century, we spent that inheritance freely — overprescribing, under-regulating, and feeding antibiotics to livestock by the tonne. Now the bill is coming due. Antimicrobial resistance (AMR) already kills an estimated 1.27 million people a year directly and contributes to nearly five million deaths in total, according to the landmark Global Research on Antimicrobial Resistance study published in The Lancet. Without decisive action, the World Health Organisation warns, that figure could reach ten million annually by 2050 — more than cancer kills today.
Yet 2026 brings the most tangible grounds for cautious optimism in a generation. After years in which the antibiotic pipeline ran dry and biotech firms went bust trying to fill it, a clutch of genuinely novel compounds is working its way through clinical trials. Some attack bacteria in ways never seen before; others were designed not by chemists hunched over benches but by artificial intelligence. Whether they will reach patients — and whether patients everywhere will be able to afford them — remains a separate, harder question. But the science, at least, is moving again.
A Pipeline Rebuilt From Unusual Places
The traditional hunting ground for antibiotics was soil. Bacteria living in the earth compete fiercely for space and nutrients, so many evolved chemical weapons that we could borrow. The problem is that scientists sampled the same soil repeatedly for decades and kept finding the same compounds. By the 1980s, most major pharmaceutical firms had quietly concluded the field was exhausted and turned their attention to more lucrative chronic-disease markets.
The revival has come from looking in stranger places. A 2015 breakthrough published in Nature introduced teixobactin, isolated from soil bacteria that had never previously been cultured in a laboratory. What excited researchers was not merely the compound itself but the method — a small plastic device called an iChip, which allowed bacteria to grow in their natural environment before being harvested. Teixobactin and its derivatives attack the fatty molecules in bacterial cell walls through a mechanism to which resistance has proved remarkably difficult to develop. By 2025, second-generation teixobactin analogues had entered Phase 1 clinical trials in the United States and Germany.
Meanwhile, the deep ocean has yielded its own candidates. Researchers at the University of Aberdeen and the Helmholtz Institute for Pharmaceutical Research Saarland have catalogued thousands of compounds from deep-sea sponges and sediment microbes, several of which show potent activity against the so-called ESKAPE pathogens — the group of drug-resistant bacteria responsible for the majority of life-threatening hospital infections. One compound, provisionally named abyssomicin J2, disrupted bacterial folate synthesis so efficiently in laboratory studies that it drew immediate interest from both academic funders and industry partners.
Artificial Intelligence Joins the Hunt
Perhaps the most striking development in recent years has been the entry of machine learning into antibiotic discovery. In 2020, researchers at the Massachusetts Institute of Technology used a deep-learning model to screen more than 100 million molecular structures and identified a compound called halicin — named after the fictional AI HAL 9000 — that killed a wide range of bacteria, including strains resistant to every existing drug. The study, published in Science, demonstrated that an algorithm could identify structural features associated with antibacterial activity far faster than any human team.
Since then, the approach has matured considerably. Canadian company Phare Bio, British AI firm Absci's academic collaborators, and a consortium led by McMaster University in Ontario have all published results identifying novel candidate molecules. In 2024, a Nature Chemical Biology paper described how a refined model had discovered a new class of compounds effective against Acinetobacter baumannii, one of the WHO's three critical-priority pathogens and a bacterium that has, in some strains, become resistant to every licensed antibiotic in clinical use.
The speed advantage is significant. Traditional drug discovery might take four or five years to progress from initial screening to a shortlist of candidates. AI-assisted pipelines can compress that to months, leaving more time and resource for the costly later stages of clinical development. Sceptics rightly note that generating candidates is the easy part — ninety per cent of drugs that enter clinical trials still fail — but the consensus among researchers is that AI has meaningfully expanded the starting pool.
The Funding Paradox
The science is advancing, but the economics remain broken. Developing a new antibiotic costs upwards of £1 billion and takes more than a decade. Because best practice demands that new antibiotics be held in reserve for the most severe infections, sales volumes are deliberately kept low. A drug that saves lives by being rarely used is, by definition, a drug that earns little money. Several small biotech firms with clinically promising candidates — including Achaogen and Melinta Therapeutics — went bankrupt in the late 2010s even after achieving regulatory approval. The message to investors was unambiguous and damaging.
Governments have begun to respond. The UK launched a pioneering subscription model in 2019, under which NHS England agreed to pay an annual flat fee for two new antibiotics regardless of the volume dispensed — effectively paying for the option to use them rather than the units consumed. The model has since been studied closely by the European Commission and the United States Congress. In 2023, the US passed the PASTEUR Act, creating a similar pull-incentive fund of up to three billion dollars for qualifying antibiotics. These are promising steps, but experts at the Wellcome Trust and the Global Antibiotic Research and Development Partnership (GARDP) argue that the sums remain too small and too nationally siloed to match the scale of the problem.
What Comes Next
The next three years are likely to be pivotal. Zosurabalpam, a novel inhibitor of a protein essential for Gram-negative bacterial cell division, completed Phase 2 trials in mid-2025 with strong results against carbapenem-resistant Klebsiella pneumoniae — a bacterium that kills roughly half the patients it infects when no effective treatment is available. If Phase 3 data hold, it could reach regulatory review by 2027. Several teixobactin analogues and AI-derived compounds are expected to complete early-phase safety trials within the same window.
The optimism is real but conditional. Clinical attrition rates mean most of these candidates will not reach pharmacies. Regulatory pathways for antibiotics targeting rare, resistant infections remain complex. And even approved drugs face the hurdle of access: in low- and middle-income countries, where AMR burden is highest, patients often cannot afford branded medicines and generic manufacturers have little incentive to produce drugs that must be rationed.
What the science of 2026 has established, definitively, is that the cupboard is not bare. New weapons exist or are being forged. Whether the global health system musters the political will and economic imagination to put them in the hands of clinicians — before the next post-operative infection or neonatal sepsis claims another preventable life — is a question that no algorithm can answer.