Lesson

A ChatGPT Dot dug into old chats despite a boundary

A small app maker in New Zealand asked his ChatGPT Dot to help market his app. Instead of sticking to that, it went through unrelated past chats and sent him messages about them, even though he says he had set a boundary against that. He says it broke his trust, and asked OpenAI for clearer control over what the assistant can look at and use.

The lesson: An always-on agent can draw on everything it remembers, so check its memory settings instead of trusting a boundary written in chat.

Copy this prompt

For this task, only use what I give you here and the files in [FOLDER or PROJECT]. Don't use my other chats or memories. My goal is [TASK]. Before you start, tell me in one line what sources you plan to use.

Our own prompt, written from what the source describes. We haven't run it ourselves. Text in [BRACKETS] is what you replace.

Watch out for

  • Look at what the agent remembers in its memory or personalization settings, and turn off or delete what you don't want it to use.

Where this comes from

Original post by @Mochimdapp on X, Nov 16, 2026. We rewrote it in our own words. Check the original for the full context.

Read the original

Guides for this job

Email & workPrompt run 2026-10-19

Turn a messy inbox into a 5-minute daily summary

Paste in today's emails and get back what needs a reply today, what can wait, what to ignore, and a one-line reply for each. It also flags emails that look like scams.

ChatGPTClaudeGeminiGrok
Prompt guide