Dognosis pairs trained dogs with an AI reader to prescreen breath samples for more than 20 cancers

One of the more promising cancer-screening interfaces of 2026 may be a dog’s nose with an AI translator attached. Bengaluru startup Dognosis is running a Phase 3 trial on exactly that.

How the test works

Cancers push volatile organic compounds into a person’s breath, and dogs can be trained to detect them. Dognosis builds on that with a pipeline designed to remove the handler from the loop. Patients breathe into masks. Dogs sniff the samples. Sensors track the dogs’ movement, respiration and body language, and an AI model translates those signals into standardized results, instead of relying on a human handler’s interpretation of what the dog is indicating.

The AI layer is what turns a well-known but hard-to-scale observation, that dogs can smell cancer, into something that can be run consistently across many samples and sites.

The numbers so far

From the published Phase 2 study: seven trained dogs identified more than 90% of cancers and more than 91% of non-cancer samples across seven cancer groups covering 20-plus cancer types, with similar performance on early-stage disease.

A Phase 3 trial started in April across 10 Indian hospitals, aiming to enroll roughly 10,000 people. The long-term plan is about 30 dogs supporting up to a million tests a year, which gives a sense of the throughput the company believes the sensor-and-model layer makes possible.

What it is, and what it is not

The honest caveat comes from the company’s own framing: this is a prescreening tool. A positive result routes a person to real diagnostics; it is not a diagnosis, and it does not replace the diagnostic workup that confirms cancer.

Read that way, the bar it has to clear is different from the one a diagnostic test faces. As a cheap, non-invasive first filter for more than 20 cancers at once, the comparison is not to the confirmatory diagnostics that follow. It is to what most people get today, which is nothing at all. The Phase 3 results across 10 hospitals and roughly 10,000 participants will show whether the Phase 2 detection rates hold at that scale, and that is the number to watch before the million-tests-a-year plan means anything.