Disconnecting an AI Assistant Doesn't Mean Deleting Its Data

We spend a lot of time thinking about who controls our money and infrastructure. I think we should ask the same questions about the AI services we connect to our email, documents and cloud accounts.

I went through the official documentation for ChatGPT, Claude and Gemini to understand what actually happens when we disconnect an integration.

A few distinctions stood out:

  • Revoking access is different from changing tool permissions.
  • Disconnecting a source doesn’t necessarily erase information already stored in conversations.
  • Third-party connectors can introduce additional trust boundaries.
  • Deletion and retention need to be checked separately from account access.

This is a documentation-based investigation, not a claim that I’ve tested the vendors’ internal systems.

I published the findings on Shadowbip. The free preview covers the main differences; the paid section includes an evidence matrix, an offboarding procedure and an editable AI Access Audit workbook.

https://davidebtc186.substack.com/p/you-disconnected-your-ai-assistant

Curious how other stackers handle connected AI services. Do you avoid them entirely, or use them with separate accounts and limited permissions?

https://stacker.news/items/1618032

Write a comment