nicksolarsoul floats a privacy gateway for Mike's AI calls, then pulls it

A proposal to scrub client names and organizations out of prompts before they ever reach the AI providers - opened, and closed again within a minute.

securityinfrastructure

The idea: an optional layer that routes Mike's model calls through Hey Jude, a gateway that pseudonymizes prompts - swapping out identifying details like a person's name or company - before they hit the downstream AI services. It stayed off unless an operator turned it on, and when switched on it covered all three engines Mike talks to (Claude, Gemini, and OpenAI). nicksolarsoul was upfront about the limit of the promise: Mike still keeps the original, un-scrubbed chat text in its own database, so this only shields what goes out to the providers, not what sits at rest.

None of it landed. The pull request was withdrawn about a minute after it opened, so this is a signal of intent rather than a shipped feature.

So what Firms weighing whether AI vendors ever see identifiable client data should watch this thread - it's the shape of a real answer, even if this attempt didn't stick.

View this fork on GitHub →

Spotted something wrong? Or know the PR text has fresher detail than the writeup above?