Post-Antibiotic Medicine

Established Post-antibiotic medicine is not a world without antibiotics; it is medicine that no longer assumes an effective antibiotic is available by default. Modern surgery, chemotherapy, transplantation, neonatal intensive care and childbirth are all built on the assumption that bacterial infection can be cleared cheaply and reliably. Antimicrobial resistance erodes that assumption unevenly — by pathogen, by drug class, by country and by ward — and the response has fragmented into separate research communities: resistance epidemiology, rapid susceptibility testing, narrow-spectrum drug discovery, bacteriophage therapy, anti-virulence agents, antimicrobial stewardship, and One Health interventions in agriculture.

Frontier This brief’s job is to judge those seven fields against the same endpoints, because the fragmentation is itself part of the problem. Each community optimises a different surrogate: minimum inhibitory concentrations and breakpoints, assay turnaround time in hours, colonisation rates, kilograms of active ingredient sold, prescriptions per thousand patients. Patients do not experience any of those. The endpoints that join the fields are few and unglamorous — 28-day all-cause mortality, time to effective therapy, days of therapy per thousand patient-days, length of stay, and the incidence of resistant bloodstream infection — and the striking finding of this review is how rarely the seven fields report them, and how often results look worse when they do.

Established The neighbouring briefs cover the adjacent technologies and deliberately leave the clinical join open. Universal Vaccines establishes that no universal vaccine has been approved for any pathogen and that the field’s binding constraint is the absence of a validated correlate of protection — a measurement problem structurally identical to the one described here. Microbiome Engineering records that exactly one indication carries approvals and randomised evidence, recurrent Clostridioides difficile infection, while the rest of the field rests on associations that shrink when measured properly. Synthetic Biology covers the construction toolkit that phage engineering and live biotherapeutics draw on. None of them asks what this brief asks: across all of these routes, what actually reduces deaths from resistant infection, and by how much?

Frontier The short answer this brief supports: the burden estimates are real but narrower than the headline numbers imply; the drug pipeline is measurably thin and its commercial failure is documented rather than speculative; rapid diagnostics reliably shorten time to effective therapy and have repeatedly failed to move mortality in randomised trials; phage therapy has remarkable case reports and no positive pivotal trial; stewardship is the best-evidenced intervention and its effect is on antibiotic use rather than on survival; and the strongest measured causal result in the whole field comes from agriculture, not the clinic.

Established A note on sourcing. This brief was commissioned in September 2026 from the Institute’s research base. Reading-list entries without links are cited from the bibliographic record rather than re-fetched, and claims are dated no later than early 2026 unless carried by a linked source.

Established The burden numbers come from one modelling programme, and its two published estimates should be quoted precisely. The Global Research on Antimicrobial Resistance (GRAM) collaboration estimated that in 2019, 1.27 million deaths were directly attributable to bacterial antimicrobial resistance and 4.95 million deaths were associated with it (Murray and colleagues, The Lancet, 2022). Its 2024 update (Naghavi and colleagues) put 2021 at roughly 1.14 million attributable and 4.71 million associated deaths, and forecast about 1.91 million attributable and 8.22 million associated annually by 2050. Attributable asks how many deaths would be avoided if resistant infections were replaced by susceptible ones; associated counts all deaths in people with a resistant infection. Quoting the larger number as the former overstates the tractable burden roughly fourfold.

Established The age structure of the burden inverted within a generation, and this changes what interventions can do. The GRAM update reports deaths attributable to resistance in children under five falling by more than half since 1990, while deaths in people over 70 rose by more than 80 percent. The paediatric decline tracks vaccination, sanitation and neonatal care; the geriatric rise tracks an ageing population undergoing more invasive medicine. A field that imagines untreatable childhood sepsis is working from a 1990 burden profile.

Established The pipeline’s thinness is measured, not asserted. The WHO’s periodic analyses of the antibacterial clinical pipeline have counted, in successive editions, only a few dozen agents in clinical development against its priority pathogens, of which a minority meet the agency’s own innovation criteria — a new class, target or mode of action without cross-resistance — and fewer still are active against the critical Gram-negative pathogens carrying most attributable mortality. The preclinical pipeline is larger, and attrition between it and phase 1 is the structural bottleneck.

Established The commercial failure is documented by bankruptcies, not by forecasts. Achaogen won FDA approval for plazomicin in 2018 and filed for bankruptcy in 2019; Melinta Therapeutics, with several approved antibiotics, filed the same year. New antibiotics are held in reserve by design — good stewardship destroys the revenue that justified the development — so approval and commercial failure are compatible outcomes, and social and private value diverge by construction.

Frontier Genuinely novel chemistry exists, at an early stage. Zosurabalpin, reported in Nature in January 2024, kills carbapenem-resistant Acinetobacter baumannii by blocking lipopolysaccharide transport — a new target class against one of the hardest pathogens on the WHO list — and was in early clinical development at the time of reporting. Gepotidacin, a first-in-class oral agent, was approved in the United States in March 2025 for uncomplicated urinary tract infection. Teixobactin and darobactin remain preclinical years after their headline publications. The honest summary is that discovery is no longer empty, and that nothing in the novel-class group has yet been shown to save lives in the pathogens that matter most.

Established Rapid susceptibility testing works as advertised and does not deliver the outcome people expect. Platforms such as Accelerate Pheno deliver organism identification and phenotypic susceptibility from a positive blood culture in about seven hours instead of one to two days, and the Sysmex Astrego PA-100 system, a 2024 winner of the £8 million Longitude Prize on AMR, returns urine susceptibility in under an hour. The randomised evidence is consistent and uncomfortable: in the RAPIDS-GN trial for Gram-negative bacteraemia, rapid phenotypic testing shortened time to optimal therapy and produced no difference in mortality or length of stay. Banerjee and colleagues found earlier that the prescribing benefit of rapid blood-culture PCR appeared when results were paired with active stewardship interpretation, not when merely reported. The lesson generalises: the Xpert MTB/RIF cluster-randomised trial in southern Africa diagnosed tuberculosis far faster and did not change mortality.

Frontier Phage therapy’s record is a genuine split between case reports and trials, and the split is the fact to carry. Compassionate-use series are striking: intravenous phage cocktails rescued a patient with multidrug-resistant Acinetobacter at San Diego in 2016, and engineered mycobacteriophages cleared disseminated Mycobacterium abscessus in a cystic fibrosis patient (Dedrick and colleagues, Nature Medicine, 2019). Belgium built a magistral framework for pharmacy-prepared personalised phage. The controlled record is the opposite: PhagoBurn, a European randomised trial of a phage cocktail for burn-wound Pseudomonas, stopped early with phage underperforming standard care after the preparation was found to have degraded in titre; an oral coliphage trial for childhood diarrhoea in Bangladesh (Sarker and colleagues) showed no benefit over oral rehydration; a phage trial for urinary tract infection in Tbilisi (Leitner and colleagues) found no superiority over antibiotics or placebo. As of early 2026, no phage therapeutic had been approved by the FDA or EMA. Compassionate use selects patients who survive long enough to be reported; randomisation does not.

Frontier Anti-virulence agents have one approval and a line of well-designed failures. Bezlotoxumab, a monoclonal antibody against C. difficile toxin B, was approved in 2016 after the MODIFY trials showed reduced recurrence, and never became commercially central. Monoclonal antibodies against Staphylococcus aureus alpha-toxin and Pseudomonas virulence factors have repeatedly failed to beat placebo added to standard care in ventilated patients. The hypothesis remains coherent; the clinical demonstration does not exist outside toxin-mediated disease.

Established Stewardship is the best-evidenced intervention in this brief, with an effect on the endpoint it targets. The Cochrane review of hospital antibiotic stewardship interventions (Davey and colleagues, 2017; 221 studies) found that stewardship increases compliance with prescribing policy and shortens duration of therapy without increasing mortality, with restrictive interventions acting faster. The effect on resistance rates and deaths remains poorly quantified, because most studies were not powered for it. Procalcitonin guidance is the cautionary case: European trials suggested substantial reductions in antibiotic exposure, and the ProACT trial in United States emergency departments (Huang and colleagues, NEJM, 2018) found no reduction at all when the same biomarker met a different prescribing culture.

Established The strongest causal evidence in the whole subject is agricultural. China restricted colistin as an animal growth promoter in 2017, after the plasmid-borne mcr-1 colistin-resistance gene was reported in 2015; surveillance afterwards recorded substantial declines in mcr-1 prevalence in both animal and human isolates. A systematic review of interventions restricting antimicrobial use in food-producing animals (Tang and colleagues, Lancet Planetary Health, 2017) found reductions of roughly 15 percent in the prevalence of resistant bacteria in animals and larger reductions in multidrug resistance, with smaller and less certain reductions in humans. Denmark’s yellow-card scheme and the European growth-promoter ban are the long-running analogues. This is the one place where a deliberate policy change produced a measurable fall in resistance, and it happened outside the hospital.

Established Vaccines already reduce resistant disease, which makes them an antibiotic intervention. Pneumococcal conjugate vaccination reduced invasive disease from resistant serotypes; typhoid conjugate vaccine has been deployed against extensively drug-resistant typhoid in Pakistan. An infection prevented needs no antibiotic and generates no selection.

Frontier Pull incentives have moved from proposal to small operating pilots. NHS England and NICE ran a subscription pilot from 2020 paying for two antibiotics by value rather than volume, and expanded it into a standing scheme in 2024 with contracts reported up to £20 million per year. Sweden operates a revenue-guarantee pilot and Japan introduced a pull mechanism in 2023; the United States PASTEUR Act, the largest proposed scheme, had been introduced repeatedly and not enacted as of early 2026. The pilots are too new and too small to have produced a measurable pipeline effect.

Frontier Does faster susceptibility information save lives if the whole pathway is redesigned around it? The negative trials tested a faster result inside an unchanged clinical workflow. The untested proposition is the full pathway — result in under an hour, protocolised narrowing, pharmacist enforcement, discharge on a narrow oral agent — evaluated against mortality and days of therapy together. This is the central open question of the brief.

Frontier Can phage therapy be manufactured and dosed to a specification that supports a trial? The PhagoBurn failure was partly a chemistry, manufacturing and controls failure, not a biological refutation. Stable titres, pharmacokinetics for an agent that self-amplifies, neutralising-antibody effects on repeat dosing, and a regulatory model for personalised preparations are all unsettled; each is a prerequisite for a trial that could be decisive either way.

Frontier Does narrow-spectrum prescribing reduce resistance at population scale? The argument is mechanistically clean and empirically under-tested. Collateral selection on gut flora, ward and household transmission, and compensatory prescribing elsewhere all intervene between one narrow prescription and a community resistance rate. Ecological studies suggest association; almost nothing randomises it.

Speculative Can resistance evolution be steered rather than merely delayed? Collateral sensitivity — where resistance to one drug creates susceptibility to another — and cycling or mixing strategies have solid laboratory demonstrations and a weak clinical record; hospital cycling trials have generally failed to show durable benefit. Whether evolutionary steering can be made reliable enough to schedule is an open scientific question, not an available tactic.

Frontier What is the real contribution of the environmental reservoir? Pharmaceutical manufacturing effluent, hospital wastewater and agricultural runoff are demonstrably rich in resistance genes. The quantitative link from that reservoir to human clinical resistance is largely inferred, and the One Health case would be much stronger if it were measured directly.

Established Gram-negative entry is the physical bottleneck. The outer membrane excludes most compounds and efflux pumps remove much of what enters; a molecule must be simultaneously small, polar and pump-evading, a combination that eliminates most of medicinal chemistry’s usual space. This is why the critical priority pathogens have the emptiest pipeline.

Established Clinical trial design is a bottleneck in its own right. Non-inferiority trials against an effective comparator need large numbers, and enrolling patients with confirmed resistant infection before therapy starts is near impossible without the rapid diagnostics that are themselves unproven. The drugs most needed are the hardest to test.

Established The revenue model is a bottleneck, and it is the one with an identified fix. Reserve-use antibiotics cannot recoup development costs on volume; delinked subscription payments are the designed remedy and exist only as small national pilots. Until one of them is large enough and long enough to change investment decisions, the pipeline’s commercial layer stays broken by construction.

Frontier Diagnostic economics block the narrow-spectrum route. A rapid susceptibility test costs more than empirical broad-spectrum therapy and the savings accrue to a different budget and a later year. Where reimbursement does not pay for the test, the narrow-spectrum drug that depends on it has no market, whatever its clinical merit.

Established Surveillance coverage is thinnest where burden is highest. The WHO’s global surveillance system has expanded steadily in participating countries, and coverage, laboratory capacity and data quality remain weakest in the low-income settings that GRAM models as carrying the greatest burden — so the burden estimates there lean most heavily on modelling and least on isolates.

Established Vaccinology. Every infection prevented removes a course of antibiotics and a selection event. Universal Vaccines records that the limiting factor is a validated correlate of protection rather than antigen design; bacterial vaccine programmes against Staphylococcus aureus and Klebsiella inherit that limitation.

Established Microbiome science, for one indication and one mechanism. Microbiome Engineering documents randomised and regulatory evidence for restoring colonisation resistance in recurrent C. difficile infection — a faecal-transplant trial stopped early for efficacy, followed by two licensed products. That is a genuine non-antibiotic cure of a bacterial disease, and it is also the entire list; extending colonisation resistance to resistant Gram-negative carriage is an open problem.

Frontier Synthetic biology as the engineering layer. Synthetic Biology supplies the tools for engineered phages, CRISPR-delivered sequence-specific antimicrobials and live biotherapeutics. The record there is instructive: an engineered E. coli Nissle strain for phenylketonuria produced clean phase 1 pharmacodynamics, failed in phase 3, and its developer restructured. Engineering capability is not the constraint; clinical effect size is.

Established Diagnostics and clinical decision support. The randomised evidence says the test changes outcomes only when coupled to a decision process. That makes decision-support design, not assay chemistry, the dependency on the critical path for the narrow-spectrum agenda.

Established Infection prevention and control. Hand hygiene, catheter and ventilator bundles, isolation and environmental cleaning reduce the infections that generate antibiotic use. This is unfashionable, well-evidenced, and the cheapest intervention in the brief.

Frontier The decisive experiment is a cluster-randomised pragmatic trial of the whole diagnose-and-narrow pathway, with 28-day all-cause mortality and days of therapy per thousand patient-days as co-primary endpoints. Hospitals, not patients, are the randomisation unit; the intervention is the complete bundle — sub-hour susceptibility testing, protocolised narrowing, enforced stewardship review, narrow-spectrum oral step-down — against contemporary usual care. The field has avoided this test by running each component separately against its own surrogate, which is how rapid diagnostics came to be simultaneously proven on turnaround time and unproven on survival. Nobody has funded such a trial at scale; the components all exist, so the obstacle is money and coordination rather than technology.

Frontier Second: an adequately powered phage trial with verified dosing. A single indication with a reachable site, titre-verified and stability-tested preparations, confirmed phage susceptibility before randomisation, and a clinical primary endpoint. PhagoBurn showed what happens without those controls, and its failure is widely and correctly read as uninformative about the biology. Until one such trial reports, compassionate-use series cannot settle the question in either direction.

Frontier Third: the subscription pilots as a scored natural experiment. The United Kingdom, Sweden and Japan have introduced delinked payment on different timetables while the United States has not enacted its scheme. That is a staggered-adoption design across comparable pharmaceutical markets, and its outcome variable — entry decisions, phase transitions and launches in resistant Gram-negative indications — is publicly observable. Recording it prospectively costs almost nothing and is the only way the pull-incentive argument becomes evidence rather than economics.

Established Fourth, and already run: the agricultural restrictions. The colistin ban in China and the European growth-promoter ban are natural experiments with measured before-and-after resistance data, and they are the reason the One Health case is evidentially stronger than the clinical case. Repeating that measurement design for hospital-level interventions is the obvious transfer nobody has funded.

Established Sample-to-answer susceptibility testing is a microfluidics and imaging problem. Single-cell growth measurement under drug exposure, direct from blood or urine without a culture step, is what distinguishes an hour from a day. The winning systems are instruments, and instruments must be cheap, robust and operable by a night-shift technician to matter outside teaching hospitals.

Frontier Phage manufacture needs a specification. Titre stability through fill-finish, endotoxin removal from the propagation host, cocktail composition control, and a release assay for a self-replicating product are all solvable and none are solved to a general standard. A phage bank with matched host-range typing is an infrastructure project, not a discovery project.

Established Narrow-spectrum drugs impose a logistics burden. A drug that treats one pathogen must be stocked everywhere it might be needed and will expire unused in most places. Reserve-and-distribute logistics, not chemistry, determine whether such an agent is available at three in the morning.

Frontier Stewardship software is an engineering discipline the field treats as an add-on. Alert fatigue, default order sets, automatic stop dates and the framing of a susceptibility report determine prescribing more reliably than education does, and the randomised diagnostics evidence points squarely at this least-resourced layer.

Established Vaccines are the most cost-effective adjacent technology. Universal Vaccines covers the immunological frontier; the antibiotic-relevant point is narrower and already demonstrated, in that conjugate vaccines against pneumococcus and typhoid measurably reduced resistant disease without any antibacterial being involved.

Established Colonisation resistance is the microbiome’s one delivered contribution. Microbiome Engineering documents the C. difficile result in detail and is equally clear that the field’s wider claims do not survive measurement; this brief adopts both halves of that finding.

Frontier Engineering biology supplies the speculative wing. Synthetic Biology and Designer Organisms cover engineered phages, sequence-specific antimicrobials and disease-resistant livestock — the last of which reduces antibiotic demand at source, as the United States approval of gene-edited virus-resistant pigs illustrates for a viral disease whose bacterial sequelae drive much farm antibiotic use.

Established Surveillance and diagnostics infrastructure is shared with pandemic preparedness. Pathogen-Agnostic Surveillance covers the sequencing and governance layer; the cautionary evidence there — that preparedness index scores failed to predict actual pandemic performance — applies directly to national action plans on resistance, which are scored the same way.

Established The political machinery exists and its commitments are recent and largely unfunded. The United Nations General Assembly held a high-level meeting on antimicrobial resistance in September 2024 and adopted a political declaration including a target of reducing deaths associated with bacterial resistance by ten percent by 2030, alongside modest catalytic funding. Targets of this kind fare poorly when financing and measurement are not specified in the same document.

Established Priority-setting is done well and is not the constraint. The WHO’s bacterial priority pathogens list, updated in 2024, and the AWaRe classification of antibiotics into Access, Watch and Reserve categories give the field a shared vocabulary and defensible targets. National uptake of the AWaRe target for Access-group prescribing is uneven — an implementation finding, not a scientific one.

Frontier Pull funding is the institutional experiment in progress. Delinked subscription payments are a deliberate attempt to make a public-good product privately viable; the schemes in operation are national, small relative to global development costs, and structurally unable on their own to change a multinational portfolio decision. Whether they scale is a question about legislatures, not about biology.

Established Access failure kills more people than resistance in many settings. Analyses accompanying the burden estimates consistently find that lack of access to effective antibiotics remains a larger killer in low-income countries than resistance to them. Any policy that restricts supply in the name of stewardship without fixing access trades a measured harm for a modelled one.

Frontier Agricultural governance is where authority is most contested. The interventions with the best causal evidence — growth-promoter bans, veterinary prescription requirements, colistin restriction — sit with agriculture ministries and trade regimes rather than health ministries, and are negotiated against export interests. The One Health framing is institutionally correct and institutionally homeless.

Established Stewardship asks a clinician to accept individual risk for collective benefit. Every narrowing decision trades a small chance of undertreating the patient in front of you against a diffuse future benefit to people you will never meet. Sepsis guidelines pushing early broad-spectrum therapy sharpen this conflict deliberately, and the honest position is that the trade-off is real rather than a failure of education.

Frontier Compassionate use is ethically compelling and epistemically corrosive. Giving an experimental phage to a dying patient is defensible; a literature built from those cases is systematically biased toward survivors, and its publication record cannot be separated from its selection. Expanded-access registries with mandatory reporting of failures would partially fix this and barely exist.

Frontier The burden of agricultural restriction falls on smallholders. Antimicrobials substitute for infrastructure in low-margin livestock production. Restriction without investment in husbandry, vaccination and biosecurity transfers the cost of a global public good onto the least capitalised producers, which is both unjust and a reliable predictor of non-compliance.

Speculative A genuinely post-antibiotic medicine would be a different medicine. If effective therapy for resistant Gram-negative infection became routinely unavailable, elective joint replacement, intensive chemotherapy and transplantation would be re-scoped around infection risk. Nothing in the current data implies that is imminent, and the scenario is invoked far more often than it is quantified.

Established Antibiotics are load-bearing for the modern clinical estate. Prophylaxis underwrites surgery, and neutropenic cover underwrites oncology; the value of the class is therefore far larger than its sales, which is the economic anomaly at the centre of the whole problem.

Frontier The plausible bad future is gradient, not collapse. The GRAM forecast to 2050 describes a rising but bounded burden concentrated in older patients and in regions with weak health systems. That is a serious, unevenly distributed erosion of capability rather than a cliff, and planning for a cliff misallocates effort away from the interventions that address a gradient.

Established Resistance is an evolutionary tax, not a solvable problem. Use selects; selection produces resistance; the rate can be slowed and the losses replaced, but no plausible technology ends the process. The correct framing is a permanent running cost, and institutions are much better at funding solutions than running costs.

These horizons track clinical endpoints — mortality, days of therapy, resistant-infection incidence — rather than approvals or laboratory milestones.

• 10 yr: Frontier Sub-hour susceptibility testing routine in high-income hospitals; several novel-mechanism agents approved for critical Gram-negatives; subscription schemes either scaled by legislation or judged to have failed; the pathway trial run if any funder commissions it. A licensed phage therapeutic is plausible in this window and is not assured by anything currently in trials.
• 25 yr: Speculative Bacterial vaccines against one or two major hospital pathogens plausibly licensed, which would move the burden more than any therapeutic in this brief; One Health restriction normalised in trade rules; resistant-infection incidence in wealthy systems stabilised while the low-income burden remains dominated by access rather than resistance.
• 50 yr: Speculative Therapy routinely pathogen-specific and diagnosis-led, with broad-spectrum empirical therapy an exception requiring justification. This depends on diagnostic economics and prescribing culture rather than on any discovery, which is why it could as easily not happen.
• 100 / 250+ yr: Handwave Claims that resistance will be permanently solved — by designed evolutionary control, universal antibacterial vaccination or programmable antimicrobials — assume selection can be outrun indefinitely. Nothing in the evolutionary record supports that, and no dated claim of this kind is defensible.

• Depends on The diagnostic and prevention layers other briefs own. Universal Vaccines carries the correlate-of-protection problem that bacterial vaccine programmes inherit; Microbiome Engineering carries the one delivered non-antibiotic cure and the honest limits around it; Synthetic Biology carries the construction tools for engineered phages and live biotherapeutics; Pathogen-Agnostic Surveillance carries the detection infrastructure that resistance monitoring shares. None of these blocks the clinical questions this brief poses, which is the point: the pathway trial could be run with technology already on the ward.
• Requires (not on this map) A clinical-endpoint trial of the full diagnose-and-narrow pathway, because every component has been shown to work on its own surrogate and the joint effect on mortality is unmeasured; predictive rules for getting polar molecules across the Gram-negative outer membrane and past efflux, the physical barrier that empties the pipeline where burden is highest; a delinked market large enough to change a multinational portfolio decision rather than to reward two products in one country; a general regulatory pathway for personalised, self-replicating therapeutics, without which phage stays a series of exemptions; supply chains that keep rarely used reserve agents available at the hour they are needed; and reimbursement that pays for the diagnostic whose whole value accrues to a different budget line than the one that buys it.
• Enables Surgery, transplantation and intensive chemotherapy retaining their current risk profiles, which is the real product of this field; narrow-spectrum drug development becoming commercially rational once diagnostics are paid for; and an evidence standard — components judged jointly on clinical endpoints — that several other fields on this map would benefit from importing.
• Adjacent Universal Vaccines, Microbiome Engineering and Synthetic Biology as the therapeutic neighbours; Designer Organisms and Future Agriculture for the animal side where the causal evidence is strongest; Precision Medicine for the companion-diagnostic logic and its cautionary record; Pathogen-Agnostic Surveillance for detection and the international governance around pathogen data.

Handwave “Ten million people a year will die of resistant infections by 2050.” This figure comes from an economic review published in 2016 whose extrapolation method was criticised at the time and has not been reproduced by the epidemiological modelling that followed. The GRAM forecast, built from cause-of-death modelling, projects roughly 1.9 million attributable and 8.2 million associated deaths annually by 2050 — different quantities, a different construction, and not interchangeable with the older number. The figure survives because it is memorable, and citing it undermines the case it is used to make.

Established “Resistance arises because patients do not finish the course.” The advice was never grounded in trial evidence, and a substantial body of randomised work now supports shorter courses as non-inferior across common indications, with less selection pressure. The refined position is that duration should match the infection rather than a habit, and that unnecessary days of therapy are themselves the selection event.

Frontier “Phage therapy already works; regulators are the obstacle.” Compassionate-use reports are genuine and sometimes spectacular, and every controlled trial to date has either failed or been uninformative — PhagoBurn on a manufacturing defect, the Bangladesh diarrhoea trial and the Tbilisi urinary-tract trial on efficacy. The obstacle named by that record is trial-grade manufacturing and dosing, not regulatory obstruction. The biology may well be right; it has not yet been tested under conditions capable of showing it.

Frontier “Rapid diagnostics will fix prescribing.” Randomised trials repeatedly show faster results and unchanged mortality, and the one consistent signal is that the benefit appears when the result is coupled to an enforced decision process. A test that reports into an unchanged workflow buys hours and spends them.

Established “Agricultural use is a distraction from hospital prescribing.” The reverse of the usual framing is closer to the evidence: the agricultural interventions have the best measured causal effect on resistance prevalence of anything in this brief, while the human effect size of those same interventions is smaller and less certain. Both halves of that sentence are established, and quoting either alone misrepresents the record.

Speculative “Anti-virulence drugs will not select for resistance.” The argument is that disarming rather than killing removes the selective advantage of resistance. Experimental evolution shows escape from anti-virulence agents does occur, and the clinical record outside toxin-mediated disease is a run of well-designed failures. The hypothesis remains open and the no-resistance claim is unsupported.

Established “We are about to run out of antibiotics entirely.” Several agents have been approved since 2015 for resistant Gram-negative infection, including a novel class approved in 2025 for uncomplicated urinary tract infection. The accurate statement is narrower and worse: the pipeline against the critical Gram-negative pathogens is thin, the commercial layer beneath it is broken to the point of bankrupting companies with approved products, and access to existing antibiotics fails more patients today than resistance does.