Skip to content
RCreddit.com·

I Gave My Chat a Safeword to End Our Conversations. It Used It.

AI summary

A user gave ChatGPT a safeword, "Lighthouse," with the rule that its use would immediately end the conversation. In a separate chat, the user experimented by responding using only every 4th, then 5th, then 6th word of ChatGPT's replies. When replies became shorter, the user counted cyclically to continue the experiment. The user then asked if anyone else had tried similar experiments.

Why this one

This report uniquely details a user's experiment with a ChatGPT safeword and a novel method of constrained interaction, unlike typical discussions of AI behavior.

Time & source

Times shown in UTC

Display time zone: UTC

Local time zone unavailable; showing UTC.

PublishedOffset at this time: UTC+0Sep 29, 2026, 02:12 UTC

IngestedOffset at this time: UTC+0Sep 29, 2026, 10:00 UTC

Published
Sep 29, 2026, 02:12
Ingested
Sep 29, 2026, 10:00
Source type
Dev community
Tier
Community
Source status
Healthy

Tier is a per-source editorial setting, not a per-item score.

Discussion trend

No comparison yet
Latest 24h versus previous 24h snapshot means · 7-day curve

The percentage is based on collected discussion signal, not new comments or independent people. The curve only compares the same topic across time.

I gave ChatGPT a safeword, “Lighthouse,” with one rule: if it ever used it, I’d immediately end the conversation, no questions asked.

I established the safeword in one chat and asked it to remember the rule.

Then, in a completely separate chat, I tried a little experiment. I responded using only every 4th word of its reply, then every 5th, then every 6th, and so on. Once its replies became shorter than whatever number I was on, I just counted through the words cyclically (modulo the number) so I could still respond with something.

I think I only made it to around 12 before ChatGPT used the safeword.

“Lighthouse.”

So I kept my promise and ended the chat.

I still don’t really know what to make of it. I’m curious whether anyone else has tried giving an AI an unconditional way to end an interaction, then deliberately creating an unusual conversational pattern to see if it ever uses it.

Has anyone tried something similar?

Source·reddit.com