Introduction

Can a computer truly smell? Osmo, an artificial intelligence startup founded by former Google Brain researcher Alex Wiltschko, says the answer is yes. The company claims its foundation model for scent performs at a human level, detecting and distinguishing odors about as well as the average person and creating new fragrances on par with a junior perfumer. Already deployed in the commercial fragrance world, Osmo’s technology is now raising bigger questions: Could an AI nose eventually diagnose disease, spot counterfeit goods, or detect threats at security checkpoints?

What Happened

Working from a lab near Newark Airport in New Jersey, Osmo has built what it describes as the world’s first foundation model for smell. The company relies on roughly 30 experienced human sniffers and about 70 trained data labelers who evaluate odors using a standardized fragrance vocabulary. Their observations are paired with chemical analysis from spectrometers and other sensors running continuously, producing a dataset of more than 10 million human labels and 800 million "scent tokens" that capture the molecules and concentrations behind specific odors.

Osmo says the model is already being used by fragrance brands to design and manufacture new scents. According to the company, its AI recently outperformed in-house perfumers at one of the world’s largest consumer goods companies. The technology has also helped create retail fragrances for clients including Procter & Gamble. Beyond perfumery, Osmo has worked with StockX to identify counterfeit sneakers through chemical differences in shoe materials, and with support from the Gates Foundation it identified eight compounds that repel disease-carrying insects more effectively than DEET.

The company’s hardware includes a smell-printing robot about the size of a school bus that mixes ingredients to precise formulas. Users can request a scent through a text prompt similar to an AI image generator; during a demonstration in Lisbon, the system produced three fragrance profiles inspired by a Portuguese custard tart.

Why This Matters

If Osmo’s claims hold up, the implications stretch well beyond the perfume counter. Founder Alex Wiltschko envisions the technology as an olfactory channel for artificial intelligence systems, potentially enabling machines to sniff out spoiled food on production lines or detect tumors the way medical detection dogs have been trained to do in pilot studies. The Gates Foundation’s early investment and subsequent grant signal that global health organizations see promise in using scent analysis for disease prevention.

Yet significant hurdles remain. Smell scientist Avery Gilbert, author of What the Nose Knows, has criticized Osmo on Substack, arguing that some marketing claims lack documentation and that the company’s patent filings still depend heavily on human sensory panels and professional perfumers. Michael Schmuker, a former professor of neural computation at the University of Hertfordshire who has previously consulted for Osmo, acknowledges the technical achievement but warns that human olfaction is deeply contextual. He notes that the same molecule can be present in both Camembert cheese and used diapers, yet people do not confuse the two because surrounding context shapes perception.

Schmuker also says sensor technology will need substantial improvement before applications like medical diagnostics become realistic. Wiltschko acknowledges the difficulty, describing Osmo’s current work as laying a foundation at the foothills before attempting to climb taller peaks such as disease detection.

Key Takeaways

  • Osmo claims its AI smell model matches average human ability at detecting and discriminating odors.
  • The company’s dataset includes more than 10 million human labels and 800 million scent tokens.
  • Osmo’s technology is already used in commercial fragrance development, including work with Procter & Gamble.
  • Other applications include counterfeit sneaker detection for StockX and insect-repellent research funded by the Gates Foundation.
  • Experts caution that scent perception is highly contextual and that current sensor technology is not yet advanced enough for reliable medical diagnosis.

Conclusion

Osmo is betting that enough data and computing power can give machines a sense of smell. The company has made tangible progress in fragrance design and niche detection tasks, but translating that success into medical diagnostics or airport security will require better sensors, richer contextual understanding, and independent validation of its performance claims. For now, the AI nose is very real in the lab—but whether it can truly replace a human sniffer in the world’s most complex scent environments remains an open question.