Trade secret protection requires you to point at a thing and name it. The most valuable assets in an AI-driven business are not things. They are processes, and the law has no vocabulary for them yet.
Every trade secret claim begins with the same demand, and it is the one most companies cannot satisfy. Identify the secret. Not the category. Not the system. The specific information, described with enough particularity that a court can tell whether it was taken.
For a customer list, this is trivial. For a chemical formula, a source code module, a pricing model, it is manageable. For the thing that actually makes an AI-driven business valuable, it is close to impossible, and the impossibility is structural rather than a failure of drafting.
The Identification Problem
Consider what a competitor gains by hiring your senior machine learning engineer. Not a file. Possibly nothing that would appear on any exfiltration report, and possibly nothing that engineer could hand over if asked.
What they gain is the knowledge of which four approaches were tried and abandoned, and why. The intuition for which data sources were noisy. The awareness that the obvious architecture fails at a particular scale for a non-obvious reason. The judgment about which evaluation metric actually predicts production performance. That knowledge is worth more than any artifact in your repository; it took three years and substantial capital to generate, and it is nowhere.
It is not in a document. It is not on a device. It is not recoverable by forensics, because it was never written down. It walked out in a head and the law has no clean way to reach it.
Negative know-how is recognized in trade secret doctrine, at least in principle. In practice, asserting that a competitor benefited from knowing what does not work requires you to describe the failed approaches with specificity, which means disclosing them, which means teaching the competitor the very thing you are trying to protect. The remedy and the harm are the same act.
Where This Bites Hardest
This is not a problem confined to AI companies. It applies wherever value has migrated from artifacts to process, which is most places.
A pricing team that has run four years of experiments knows the shape of demand in a way no model file captures. A drug discovery group knows that targets are dead ends. A procurement function knows which suppliers actually deliver against which terms, information that exists in nobody’s system and everybody’s judgment. In each case, the recorded artifact is a shadow of the real asset, and it is the shadow the law is equipped to protect.
AI has widened this gap rather than created it. Tools now let a departing employee reconstruct from memory what previously required documents. Describe the architecture, the data characteristics, the failure modes, and a model will regenerate a working approximation. The employee took nothing. The competitor has it anyway. There is no exfiltration event to detect, no artifact to trace, no chain of custody to establish, and your forensic capability, however good, is looking for the wrong thing.
What Follows From This
The honest conclusion is uncomfortable. Trade secret litigation is becoming a weaker instrument for the assets that matter most, and companies over-relying on it are relying on a remedy that does not reach their exposure.
The practical response is to shift weight away from litigation and toward two things that actually work.
The first is documentation as a defensive act. If negative know-how is your asset, write it down. Not for the engineers, who already know it, but for the record. A documented decision log describing what was tried and rejected converts tacit knowledge into an identifiable secret, which is the only form the law can protect. Most organizations resist this because it feels like overhead. It is the difference between having a claim and having a grievance.
The second is retention. When the asset is in a person, the control is the person staying. That is a compensation and culture question rather than a legal one, and it sits outside the remit of everyone usually asked to solve trade secret risk. It is also, in an AI-driven business, the highest-leverage control available, and it is almost never on the list.
The Litigation Posture That Remains
Where a claim is viable, the requirement is specificity before assertion rather than after. Courts have clearly signalled that they will not accept a category as a description. That means:
- Identify the crown jewels as discrete, describable assets before any dispute exists. Doing this under litigation pressure produces a definition drafted to fit the facts, which is visible and does not survive.
- Preserve early and broadly. Devices, cloud logs, repository history, collaboration platforms, file transfer records, and AI interaction histories where lawful. The window before routine deletion is short, and it closes without warning.
- Be honest internally about what the evidence supports. The distance between believing information was taken and being able to prove it is where most trade secret matters die, usually expensively, and usually after the decision to litigate has become difficult to reverse.
The Uncomfortable Summary
The gap between what is valuable and what is protectable is widening, and it is widening fastest in exactly the businesses that most need protection. Legal instruments are not going to close it, because the instruments require a noun and the asset is a verb.
The companies that handle this well will be the ones that stop asking their lawyers to solve it and start asking whether the knowledge that matters has been written down and whether the people who hold it have a reason to stay. Those are unglamorous questions. They are also the only ones with answers.
Ankura conducts trade secret risk assessment, departing-employee investigation, data-exfiltration analysis, repository and cloud review, and expert support in disputes.
© Copyright 2026. The views expressed herein are those of the author(s) and not necessarily the views of Ankura Consulting Group, LLC, its management, its subsidiaries, its affiliates, or its other professionals. Ankura is not a law firm and cannot provide legal advice.
