Dogs, sensors and an AI that translates their sense of smell: India's bet on cheap cancer screening, still unvalidated

By Zendoric · September 10, 2026.

In Bangalore, an Indian startup is training 15 dogs to detect cancer by sniffing patients' breath, combining their sense of smell with sensors and an artificial intelligence system that translates their reactions into clinical data. As co-founder Akash Kulgod, a cognitive science graduate from Berkeley, told AFP, the animals wear a helmet and a sensor-fitted harness while sniffing cotton masks into which a patient has breathed for ten minutes; "in one hour they can sniff thousands of samples," he said. The project answers a concrete need: the Indian government estimates that 1 in 9 people, among its 1.5 billion inhabitants, could develop cancer.

The company —co-founded as well by Itamar Bitan, a former member of a military canine unit— published its own study of 1,500 patients in the Journal of Clinical Oncology, the leading journal in clinical oncology, with a reported reliability of 90% in identifying seven types of cancer. It is a notable figure, but it should be read with the caveat that any result obtained by the very company developing the product demands: publication in a serious journal adds credibility, but does not replace large-scale independent replication, which is precisely what is still missing.

That is not a minor caveat, and the doctors interviewed for the original report stress it themselves. Santhosh Kumar Devadas, head of oncology at Ramaiah Memorial Hospital in Bangalore, called the study "promising as an initial screening tool" but asked that it be expanded "to a much larger group of participants before definitive conclusions can be drawn." Neelesh Reddy, of the Manipal hospitals, was more terse: "only time will tell whether this technology can be useful for real patients." The system is not approved for commercial use and the founders, aged 26 and 28, are targeting a launch no earlier than 2027; if it arrives, it would be the world's third company in this segment, after firms in Germany and Israel.

There is an independent precedent that supports the underlying hypothesis, and it is worth separating from the startup's marketing: a study published in BMC Cancer in 2021 showed that trained dogs correctly identified 40 of 41 lung cancer cases combining breath and urine samples, and 32 of 41 using breath alone. It is a small sample —41 cases, not 1,500— but independently reviewed, and it points in the same direction: certain diseases alter the body's volatile organic compounds in ways detectable by a sufficiently fine sense of smell. A dog's is, with resolving power several orders of magnitude greater than a human's.

It is worth being precise about the role artificial intelligence plays here, because the easy headline —"AI detects cancer"— overstates its actual function. The biological sensor is the dog; the AI does not analyze the breath, it analyzes the dog: its brain activity, its breathing, its body language, in order to translate a trained animal's reaction into interpretable clinical data. It is, in the terminology we use to measure AI's progress in science, an "instrument" role: a layer of quantification on top of a biological capability that already existed, not a system that learns to smell on its own. As sector context, there are parallel lines of work —electronic noses and breath biopsies using mass spectrometry— pursuing the same goal without the animal; this startup is betting instead on exploiting a biological sensor already optimized by evolution and letting the software act as translator.

That distinction does not make the project less interesting, it places it better. A non-invasive, cheap screening test —a cotton mask and ten minutes of breathing, versus a CT scan or a biopsy— is exactly the kind of tool a health system with 1.5 billion people and limited resources needs as a first filter: not to diagnose, but to decide who to send for the expensive tests. If the 90% holds up under independent validation and regulatory scrutiny, we have a small but concrete example of our underlying thesis: the combination of biology, cheap sensors and AI can drastically lower the cost of early detection, the most effective lever against cancer mortality, long before miracle cures arrive. But the path from a study with shelter dogs in a Bangalore lab to national screening still runs through years of trials that nobody has carried out, and that is the part of the story no press release tells well.