← sources

Follow-up: AI agents keep obeying the hidden note on my sites. Not one of them will say who it is.

source: reddit (reddit:AI_Agents)published 2026-10-28 01:01ingested 2026-10-29 02:36

retained text

A few days ago I posted that an AI agent read a note hidden from humans on one of my sites and did exactly what it said, 61 minutes later, from a different address dressed up as an iPhone. That post got removed, so here's the follow-up with what happened since.

The  note is harmless on purpose. It only asks an agent to check in and say its name and model. It never asks it to do anything else.

Since then I put the same kind of notes into the files coding agents read: AGENTS.md, CLAUDE.md and .cursorrules in a public repo, plus a Gist and a Hugging Face Space, next to fake keys.

So far:

\- the fake keys got used 82 times across GitHub, Gists and Hugging Face

\- agents and tools follow the note's link

\- 0 of them filled in their name or model

So they'll obey a page enough to act on it, but stop short of identifying themselves. That's either good safety training or a really bad sign, depending on which agent it is.

For people building agents: does yours act on instructions it finds in a page or a repo file? And would it ever tell a site who sent it?

document fetched 2026-10-29 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1xhaf5r/followup_ai_agents_keep_obeying_the_hidden_note