The Great and Powerful Os(mo)

Pay no attention to the human sensory panelists behind the AI

In a January 27, 2026 press release, Osmo Labs trumpeted (while providing no numbers) that it had “published” more fragrance chemical patents in 2025 than the rest of the F&F industry combined. I suggested here that Osmo was bragging about published patent applications, not grants. The difference between the two is the difference between “we’ve got a recipe idea” and “dinner is ready.”

According to Osmo, this “milestone” enables them to

explore and analyze billions of potential molecules and identify entirely new ingredient families that traditional methods cannot access, with more precision and speed.

They contrast this with the industry’s “time-consuming and expensive discovery process” where “chemists create molecules one at a time in the lab, test them with perfumers, and begin again.”

While Osmo received no actual patents in 2025 it did, as I’ve since found out, have five patent applications published. I’ve had a chance to review them and compare the actual filings to the company’s PR hype.

Fragrance and flavor companies seek patent protection for a new molecule when they are convinced it is safe, commercially feasible, and likely to be sought out by perfumers and flavorists. The new item should smell and/or perform better than existing alternatives. For example, it could deliver a more complex aroma or last longer on fabric.

The typical fragrance chemical patent describes the odor of the claimed molecule in a single sentence. For example, regarding 6-isobutyl-2-methyl-2,3-dihydro-1H-inden-1-one Givaudan wrote: “Odor description: floral fruity fatty, creamy lactonic like, nectaryl-like, rosy, rose petal, green rose oxide-like.”¹

Similarly, in its claim for novel hydroxy-4,8-dimethyl-dec-4-enal compounds, IFF stated: “The compounds of the present invention exhibit floral, sweet juicy, green and watery lily of the valley notes, which are surprisingly and unexpectedly strong and long-lasting.”²

Simple. Straight to the point. Fragrance chemical patents rarely, if ever, mention that the odor description was generated by a perfumer or fragrance chemist, much less a “Master Perfumer.” Nor do they provide a long disquisition about the training of odor evaluation panelists, the rating methodology they used, and the descriptions they arrived at when characterizing the molecules in question. It is remarkable then, that four of the five Osmo patent applications include such accounts. This heavy reliance on human sensory judgments seems odd for a company premised on super-cool AI. What’s going on?

Osmo’s core idea was to create an AI-based method that could predict the odor properties—and ultimately other properties, like toxicity—of any molecular structure fed into it. They filed a patent application³ for this method in November, 2021. Note: this application is about AI methods, not specific aroma molecules.

Like all such models, the Osmo AI smell predictor had to be trained, which meant feeding it a list of molecules along with associated odor descriptions. Osmo created the necessary dataset by running a grab bag of 400 commercially available chemical compounds past a panel of trained sniffers. It described this process in a patent filing three years later⁴:

As part of the evaluation and validation of a novel artificial-intelligence driven software model to map molecular structures to odor perception, a set of 400 commercially available chemical compounds of diverse structure were obtained and submitted to analysis by the model to generate a predicted odor profile on each compound. A cohort of subjects were also trained to describe their perception of odorants using the Rate-All-That-Apply method (RATA) and a 55-word odor lexicon . . .

According to the filing, the majority of these open-market molecules had describable, and possibly useful, odors. At this point, Osmo descended from the clouds and claimed fragrance chemical rights to the enormous and motley assemblage of molecules it had used to train the software model. Note that these molecules were not discovered by Osmo’s AI—they were used to train it.

The basis of Osmo’s claims were the olfactory ratings of their panelists, plus the judgments of a Master Perfumer and fragrance chemists as to the compounds’ commercial prospects (graded “A” through “D”). That sounds a bit like the old-fashioned “time-consuming and expensive discovery process” that smell AI was supposed to revolutionize, no?

Subsequently, the company filed four more fragrance chemical patents using the same general scheme—olfactory ratings by a trained panel. These applications did not take the spray-and-pray shotgun approach of the “FRAGRANCE AND FLAVOR COMPOSITIONS” filing. Instead they went the more traditional route of claiming multiple variations of a basic molecular structure. Specifically, these filings focused on 1-ethynyl-4-pentylbenze and related compounds,⁵ thienyl alkanoates⁶, benzine derivatives⁷, and norbornane and norbornene derivatives.⁸

While more focused in claims, the later filings provide odor notes for an extraordinary number of compounds, e.g., 127 thienyl alkanoates, 61 benzine derivatives, and 57 norbornane and norbornene derivatives. Each application is supported by odor descriptions from a sensory panel and Master Perfumer, along with odor intensity ratings based on dipped perfume blotters evaluated at multiple time points.

None of these four subsequent filings mention Osmo’s AI smell mapping software model. Perhaps AI was used to generate the lists of molecular variants. Or perhaps the lists were generated by in-house chemists, a few summer interns, or a bunch of rhesus monkeys playing with stick models. In any case, it makes no difference for the purposes of a patent application.

These filings reference only the molecules thought worthy of a patent claims—we are not told how many compounds were “created one at a time in the lab” only to be panel-tested and found to be non-smelly or bad smelling and thus not fit for filing. Thus the Osmo PR claim of “10x higher success rates compared to industry norms” remains undocumented.

All of this raises questions about Osmo. What kind of company is it? Is it an AI-based molecular prospector? Is it an IP owner looking to license useful molecules? Is it a chemical manufacturer? Is it a fragrance compounding company aimed at the small customer niche? Is it an AI-based perfume creation play based on speed to market? Or is it some mashup of the above?

My hunch is that Osmo is making a land-grab in the fragrance chemical space, with an eye toward manufacturing or licensing new aroma molecules. Is this a viable path to revenue? It depends.

Historically, a desirable captive molecule can be a money maker, either through exclusive use in client formulations (at a big markup) or by selling into the wider market as an ingredient (IFF’s Galaxolide® is an example). This can happen if the molecule clears toxicology studies and other regulatory hurdles, if it can be manufactured economically, if it plays well with other ingredients and product bases, and if it offers significantly superior olfactory performance. So yes, if all these conditions are met, Osmo could create a revenue stream from license royalties and/or raw chemical sales. Would that satisfy the VCs who have invested $130 million in what was supposed to be an industry-disruptive startup? Good question.

A final thought: a pitch that says “hey, we’ve got a ton of all-new AI-generated chemicals for your scented products!” flies in the face of current market trends toward “clean” fragrance formulas and sustainably produced natural compounds.

1. United States Patent 11,020,333 “Organic compounds” assigned to Givaudan SA.

2. United States Patent 12,428,609 “Organoleptic compounds” assigned to International Flavors & Fragrances Inc.

3. United States Patent Application Publication 20240021275 A1, MACHINE-LEARNED MODELS FOR SENSORY PROPERTY PREDICTION, filed November 12, 2021 by Osmo Labs PBC.

4. United States Patent Application Publication 20250075147 A1, FRAGRANCE AND FLAVOR COMPOSITIONS COMPRISING NEW FLAVOR AND FRAGRANCE INGREDIENTS, filed August 29, 2024 by Osmo Labs PBC.

5. Fragrance and flavor compositions comprising aromatic derivatives, International application WIPO/PCT, WO 2025/049824 A1, filed August 29, 2024 by Osmo Labs PBC.

6. United States Patent Application Publication 20250234910 A1, FRAGRANCE AND FLAVOR COMPOSITIONS COMPRISING THIENYL ALKANOATE DERIVATIVES, filed January 24, 2025 by Osmo Labs PBC.

7. United States Patent Application Publication 20250236810 A1, FRAGRANCE AND FLAVOR COMPOSITIONS COMPRISING BENZENE DERIVATIVES, filed January 24, 2025 by Osmo Labs PBC.

8. United States Patent Application Publication 20250313523 A1, FRAGRANCE AND FLAVOR COMPOSITIONS COMPRISING NORBORNANE AND NORBORNENE DERIVATIVES, filed April 03, 2025 by Osmo Labs PBC.