Satya Nadella published an essay on July 12, 2026 coining the "Reverse Information Paradox." The frame comes from Kenneth Arrow's 1962 observation that information's "value for the purchaser is not known until he has the information, but then he has in effect acquired it without cost." Patents answer that for the seller: public enablement in exchange for a time-limited exclusive right.
Enterprise AI runs the trade in reverse. The buyer is the one who must reveal proprietary knowledge to a third-party vendor to get useful output back. In Nadella's words, a company "essentially pay[s] for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful."
No misconduct is required. Institutional know-how escapes as what he calls exhaust: the prompts, the tool calls, and the corrections people make when the model is wrong. "Trace by trace, correction by correction, eval by eval."
Palantir CEO Alex Karp gave the commercial version on CNBC on July 1: technical buyers want to know "they own the means of production, and it's not being transferred to someone else." The essay quotes Karp, then moves the problem from vendor misconduct to a gap in the law.
Why This Matters
- Patent law converts disclosure into a right good against the world, including independent inventors. Courts treat trade secrets as property too, as Ruckelshaus v. Monsanto held. But that right reaches only misappropriation, not honest discovery, and disclosure to anyone not bound to secrecy extinguishes it. Contract can bind a vendor. Neither recreates the patent bargain.
- Nadella wants know-how to compound on the enterprise customer's side. A company's data, traces, and corrections should accumulate inside a boundary it controls, with nothing crossing to the vendor without consent. He asks for an equivalent to what patents do for an inventor, disclosure without forfeiture, but does not say what it would be.
- One disclosure can affect rights under both regimes. Most companies understand that trade secrets require secrecy. The effects on a later patent position, once unfiled, enabling technical material has moved through a vendor stack, are considered less often.
Rights like this get defined only after losses grow large enough to litigate, and that takes years. Companies can improve their position by weighing these risks with IP counsel and following consistent processes before any vendor agreement is signed and implemented.
More this week on what disclosure actually costs under trade secret doctrine, why data location is now a jurisdictional issue, and where contract stops helping.
Sources
- Satya Nadella, "Reverse Information Paradox" (July 12, 2026)
- Michael Kan, "Amid IP Theft Concerns, Microsoft CEO Floats New AI Patent Concept," PCMag (July 13, 2026)
- Alex Karp remarks, CNBC (July 1, 2026)
- Kenneth J. Arrow, "Economic Welfare and the Allocation of Resources for Invention," in The Rate and Direction of Inventive Activity (NBER 1962), at 615
Ruckelshaus v. Monsanto Co., 467 U.S. 986, 1002-03, 1011 (1984) (trade secrets have "many of the characteristics of more tangible forms of property"; "the right to exclude others is central to the very definition of the property interest"; disclosure to those "under no obligation to protect the confidentiality of the information" extinguishes it). Kewanee Oil Co. v. Bicron Corp., 416 U.S. 470, 476, 490 (1974) (trade secret law "does not offer protection against discovery by fair and honest means"; patent law "operates 'against the world'").