Apple is seeking an injunction against OpenAI in a trade secrets case, claiming former employees took confidential data and suggesting more people may be involved. Most readers see another chapter in a fight between giants.
The question almost nobody asks: who guarantees that data wasn't about customers?
Consider the concrete shape of it. A trade secret rarely lives alone in a vault. It lives in spreadsheets, in system logs, in support conversations. And in those same files you usually find the customer's name, the contract value, the usage data. When someone copies an entire repository, they don't separate what's confidential from what belongs to a third party. It all travels together.
In a well-designed system, every model runs under one principle: data doesn't leave what was authorised. The contract with the vendor doesn't guarantee that. The architecture does. It isn't paranoia, it's the cost of failure.
If data about a student paying monthly fees on a platform leaks, it's no defence that the model learned it by accident. The responsibility belongs to whoever built the system.
Apple and OpenAI can settle in court. The customer depending on a system built by someone else has no claim against either of them: they have one against the builder. A trade secret is the company's problem; a customer data leak is the builder's problem. The question left for any operation is what its models actually see of its clients' data.