在 WIRED Health 的讨论中，外科医生 Ara Darzi 指出抗生素抗药性正在成为一个加剧中的公共卫生危机。文本称全球每年因此死亡超过一百万人，另外还有接近五百万人死亡与耐药感染相关。这些感染比常见感染更难治疗、成本更高，导致住院天数更长，并同时推高医疗体系与病人的支出。Darzi 将 2026 年描述为一个“转折点”，同时提醒目前治疗仍多依赖临床医师在诊断流程缓慢且不确定下的判断。

耐药性上升主要由抗生素滥用与新药开发不足共同推动。当微生物接触到未能立即杀灭它们的抗生素剂量时，会进化出生存防御机制，使常规处方反而加速抗性并缩减可用疗法。2024 年《The Lancet》预测到 2050 年耐药感染可造成高达四千万人死亡。传统以培养为主的耐药诊断通常需 2–3 天，但在败血症等情况每延误一小时死亡风险上升 4% 到 9%。因此医师仍常在等待结果时使用经验法则做初始用药；WHO 指出 2023 年东南亚及东地中海区抗药感染比例高，一例三个中有一例耐药，在非洲则是五个中有一例。

AI 诊断有望缩小该空缺，Ara Darzi 表示 AI 系统在无额外实验室基础设施下可达 99% 以上准确率，且他强调对偏远地区尤其重要。英国 NHS 与 Google DeepMind 合作示范，该 AI 平台在 48 小时内识别出先前未知的耐药机制，而帝国学院研究者曾花约十年才理解。搭配自动化实验流程后，可全天候进行大量平行实验；深度学习可在几天内筛选数十亿分子结构，生成式 AI 亦可设计自然界不存在的化合物。然而，由于抗生素经济模式以销量为核心，大型制药商多已退出。英国自 2024 年起试行 Netflix 式订阅支付，政府按年度固定费用购买新抗生素使用权，瑞典也在测试部分去绑定模式。

In WIRED Health, surgeon Ara Darzi said antibiotic resistance is an escalating public-health crisis. The context reports more than one million deaths each year globally and nearly five million additional deaths linked to drug-resistant infections. These infections are harder and more costly to treat than typical infections, extending hospital stays and raising costs for both health systems and patients. Darzi frames 2026 as an “inflection point,” warning that treatment still often depends on clinician judgment because diagnostic workflows are slow and uncertain.

Resistance rises mainly from antibiotic misuse and low new-drug discovery. When exposure is sublethal, microbes evolve survival defenses, making routine prescribing an accelerant of resistance and shrinking viable treatment options. A 2024 The Lancet report predicts up to 40 million deaths by 2050. Traditional culture-based diagnostics usually take 2–3 days, but in sepsis each hour of delay can raise mortality by 4% to 9%. As a result, physicians still start empiric therapy while waiting for results; WHO data in this context indicate the heaviest burden in southeast Asia and the eastern Mediterranean, where one in three infections was resistant in 2023, and one in five in Africa.

AI diagnostics could reduce this gap: Ara Darzi says AI systems can reach over 99% accuracy without extra laboratory infrastructure, and he stresses the need is greatest in rural and remote settings. The UK NHS with Google DeepMind showed an AI platform that found previously unknown resistance mechanisms in 48 hours, versus about a decade by Imperial researchers. Paired with automation, models can run high-volume parallel experiments around the clock; deep learning can screen billions of molecular structures in days, while generative AI can design compounds not found in nature. Yet major firms have exited because antibiotic returns are weak under volume-based incentives. The UK has piloted a Netflix-style subscription model in 2024, where governments pay fixed yearly access fees rather than paying per use, and Sweden is testing a partially de-linked model.