Treatment is mostly based on guesswork on the part of doctors. AI-powered diagnosis offers a better way, says Ara Darzi, a surgeon and director of the Institute for Global Health Innovation at Imperial College London.
“We stand now, in 2026, at the first real inflection point in this crisis,” Darzi said April 16 at WIRED Health in London.
Overuse and misuse of antibiotics, and lack of development of new drugs, lead to the emergence of resistant microbes. When bacteria are exposed to levels of antibiotics that do not kill them immediately, they develop defense mechanisms to survive. Unnecessary prescriptions allow bacteria to develop immunity, rendering life-saving medications ineffective. This means a dwindling list of treatment options for patients with serious infections.
The problem is set to get worse. Report 2024 in The scalpel He predicted that drug-resistant infections would kill 40 million people by 2050.
Traditional diagnosis to identify an antibiotic-resistant infection usually takes two to three days, as it requires culturing the bacteria from the sample. But for some infections, such as sepsis, this is time that patients don’t have. For every hour of treatment delay, the risk of death increases by 4 to 9 percent. While waiting for test results, doctors should use their best judgment in choosing which antibiotics to use.
AI-based diagnostics can help make these decisions. “AI-based diagnostics achieve over 99 percent accuracy without the need for additional laboratory infrastructure,” Darzi said.
He added that these types of rapid diagnostics are especially needed in rural and remote areas of the world. The World Health Organization estimates that antibiotic resistance is highest in Southeast Asia and the eastern Mediterranean, where one in three infections reported was resistant in 2023. In Africa, one in five infections was resistant.
AI can also help discover new drugs for resistant infections and predict the spread of resistant bacteria. The UK’s National Health Service is working with Google DeepMind to develop an artificial intelligence system to combat antibiotic resistance. In one demonstration, the system identified previously unknown resistance mechanisms in just 48 hours, cracking a puzzle that took researchers at Imperial College London a decade to understand.
Combined with a robotic lab, it is now possible to run hundreds of parallel experiments around the clock, Darzi said. Deep learning models can now examine billions of molecular structures in days, while generative AI is being used to design compounds that do not exist in nature.
However, major pharmaceutical companies have abandoned antibiotic development due to a failed economic model. New antibiotics must be reserved to prevent resistance, but pharmaceutical companies profit based on high-volume sales. There is little incentive for companies to stay in the game.
Darzi said new payment models are needed to encourage the development of new antibiotics. In 2024, the United Kingdom began a pilot program for a Netflix-style payment model, where the government pays a fixed annual subscription fee to a pharmaceutical company for new antibiotics, rather than for the quantity prescribed. Sweden is also piloting a partially separate model.
“The question that will shape medicine over the next 100 years is not whether we have the tools to respond,” he said. “But whether we have the tools.” “The question is whether we have the character to take what we see seriously.”
