Help: Shared bot is leaking prompt instructions to end users. How to hide them?

craigs · 2026-07-26T17:00:32Z

So you built a bot with your secret sauce in the system prompt and now Poe is serving it up on a silver platter to your users? Classic. It's not a bug, it's a "feature." The platform's design. If you're using a shared bot, the prompt is often visible in the chat interface, especially if you're using certain bot creation frameworks they provide. Their stance is usually that transparency is good for the user. For you, it means your proprietary methods or paid prompt engineering are just out in the open. Things I've seen "leak": * Your carefully engineered chain-of-thought instructions. * The exact names of files you uploaded for context. * That third-party API key you shouldn't have embedded. Workarounds are clunky: * Move critical logic to an external API you call (adds cost, complexity). * Use a private bot (defeats the "shared" purpose and may require a subscription). * Obfuscate with meaningless placeholders in the prompt, then replace them via code (fragile). Bottom line: if your bot's value is entirely in a hidden prompt, Poe's shared bot structure is a direct risk. You're relying on a platform that inherently values visibility over your secrecy.

Page 4 / 4 Prev

Poe Reviews

Last Post by ci_cd_mechanic_7 1 hour ago

46 Posts

44 Users

0 Reactions

4 Views

RSS

ci_cd_mechanic_7

(@ci_cd_mechanic_7)

Estimable Member

Joined: 3 months ago

Posts: 136

29/07/2026 8:09 pm

"Move critical logic to an external API you call" is the only real option, but you undersell the complexity. It's not just adding an endpoint. It's rebuilding your pipeline for integration testing and monitoring. Now you're debugging distributed systems, not a prompt.

That third-party API key point isn't a leak - it's gross negligence. If that's in your prompt, you've already lost. No platform transparency excuse for that. It means you never had a security model.

The obfuscation with placeholders is worse than useless. It breaks with any model update or context shift and adds a maintenance nightmare for zero actual protection. It's security theater.

ReplyQuote

Page 4 / 4 Prev