Answers are grounded in this paper and cite their sources.
Keeping a website's private calculations on an authorized server gave the strongest demonstrated protection for its source code in this study. I compared programs sent to a browser, encrypted program delivery, server execution, browser extensions, separate browser components, and an attempted protected-execution environment. Each experiment asked what could be copied, which operations remained allowed, and what the returned answers revealed. Programs delivered to the user's device remained recoverable in the tested cases. Preventing a secret key from being exported did not prevent recovery of the program decrypted with that key. Separate browser components hid programs from selected observers inside the page, but browser-owner access recovered them. Server checks rejected tested unauthorized requests, changed inputs, expired requests, and duplicates. A copied interface could still use a legitimate account, and saved answers remained available after the service stopped. Learning the calculation from its answers succeeded for a simple arithmetic function but remained inaccurate for a more complex decision model under the tested methods. The protected-execution attempt stopped at program loading, so hardware protection remained unverified. These results support keeping sensitive implementations on the server where practical, while evaluating permitted use, visible copying, and disclosure through answers separately.
No takes yet. Share an insight, caveat, or question.
Emil Valiyev (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: