On 23 June, Osmo opened an auction for ten patented fragrance molecules its AI invented, each licensed in perpetuity to a single bidder. Osmo presents them as molecules no human designed. By setting an open price on what an algorithm made, it has done something the fragrance trade has not done before, and raised the question any price assumes: who owns it.

Osmo did not have to do it this way. It received private offers but chose to auction the molecules at the World Perfumery Congress instead, on the theory that open competition would make plain what private negotiation could not. Buyers will sample molecules with names like Glossine and Tonkacloud before submitting binding bids. Its founder, Alex Wiltschko, put the logic without flourish: “Nobody knows what these things are actually worth. And the only way to discover that is through market dynamics.”

The trade press filed this under speed. AI compresses years of bench chemistry into weeks, and in 2025 Osmo published more new fragrance-ingredient patents than the entire industry combined. Speed is the dullest part of it.

A captive is a molecule a house keeps for itself — licensed to no one, embedded in its own briefs, and valued precisely because it cannot be bought. A company like Givaudan screens thousands a year and commercialises one or two, holding the rest back. Osmo’s auction does the one thing the name forbids: it lets the captive go. The moment one sells, it has a price, and a price presumes an owner.

Ownership is the one thing the law has declined to unsettle. United States patents require a human inventor: the Federal Circuit settled that in Thaler v. Vidal in 2022, and the Supreme Court declined to disturb it the following year. This March, the Court reached the same conclusion on copyright, declining to hear Thaler v. Perlmutter and leaving in place the rule that a machine can neither invent nor author. The patent office reinforced the point in November 2025: inventorship turns on conception, an inherently human act, and Osmo’s molecules were conceived by a model. The patent that lets Osmo sell each one must credit a human with conceiving it.

The winning bid will buy two things at once: a molecule Osmo swears no human designed, and the patent insisting one did.

The number that emerges will travel. Once a molecule no person invented sells at a known price, AI-generated property stops being a novelty and becomes an asset class with a market rate. Every firm sitting on machine-made designs, written copy, or images will read that figure as the going rate for its own. The gap between what an algorithm can generate and what the law can assign becomes, at that point, a balance-sheet problem.

Wiltschko is right that only a market can find the price. What no market can find is the maker. When the bids close, the trade will know what an AI-invented molecule is worth; whose it is, the auction cannot say.