Extract from David Baake’s article “The Goldilocks Problem: Trade Secret Protection for AI”
As artificial intelligence (AI) continues to revolutionize the economy, courts are increasingly being asked to determine whether AI models and algorithms can be protected as trade secrets. Yet case law reveals that plaintiffs often struggle to obtain trade secret protection for their models, even for programs that contain elements that are both valuable and secret, because the plaintiff has not identified the program at the right level of particularity. The difficulty lies in describing the technology in a way that is general enough to be “understood by a lay person,” but detailed enough to separate it from “special knowledge of those persons skilled in the trade.” See Neural Magic v. Meta Platforms, 659 F. Supp. 3d 138, 166 (D. Mass. 2023).
Courts have repeatedly found that catchall terms and references to broad categories of technology (e.g., claims directed to “proprietary algorithms” or “machine learning techniques”) are insufficient because they are too vague to distinguish the claimed secret from what is common knowledge within the industry. But a highly detailed description—such as a recitation of all of the code that is claimed—may not be workable either because the fact finder may struggle to understand such granular detail and such an approach may fail to distinguish the conventional aspects of the program from those that are claimed to be secret.
