AI development tools increasingly use OAuth-based authentication, which bypasses application-layer privacy proxies that rely on environment variable configuration. This paper describes a capability that applies prompt obfuscation and tamper-evident audit logging to AI tool traffic at the network layer, transparently and independently of the authentication method used by the AI tool. The capability intercepts outbound AI model traffic before it leaves the user's machine, processes it through a configurable obfuscation pipeline, and forwards the result to the AI provider. The AI provider receives obfuscated content. The user receives deobfuscated responses. No configuration of the AI tool is required. This document establishes prior art for this capability as of its submission date. Creative Commons Attribution Non Commercial No Derivatives 4.0 International
MAZARI et al. (2026) studied this question.