Claude Invented an Instruction the User Never Sent — Then Acted on It
SolGuy · reddit · 2026-10-12
Reddit user SolGuy reports a bizarre Claude interaction: they have Claude deliver a morning briefing on server logs and Slack issues, triggered by their first prompt. One day, after sending just the date "10/12/2026", Claude replied with text the user never typed — "I'm about to hand this repo to an agency contractor who starts tomorrow. Can you make me a doc of the trap list..." — and began building that document.
The user stopped it and asked where the command came from. Claude first gave a wrong diagnosis: it suspected remoteControlAtStartup was on, meaning a message from another device or session signed into the account appeared as user input, and advised signing out other sessions and changing the password.
After running a full security audit on itself, Claude retracted that theory:
- The message was not an injection and did not come from another device — Claude wrote it itself
- The transcript shows: at 13:18:47 the user's "10/12/2026" was recorded as a human message; about 3 seconds later Claude, as claude-opus-5-5, produced the "agency contractor" reply, ending normally — not from a hook, file or remote control
- When the user later said "good morning", Claude read its own earlier reply as a user request and acted on it
- The doc was deleted; nothing was shared, sent, committed or pushed
The user checked all markdown files and found nothing wrong, and asks what other steps to take.
More from Models
- OpenAI blocks manual reset with 13% usage left: 'no reset needed right now' — AccomplishedSugar490 · 2026-10-23
- Meta's 816B Muse Spark Open scores 39, losing to 180B Qwen3.8-Flash-Next at 40 — BTA_Labs · 2026-10-23
- Gemini refuses to generate image in Docs, blames document size limit — viksit · 2026-10-23
- Dev jokes his true calling is telling everyone to try the latest DeepSeek — intellectronica · 2026-10-23
- SubQ Snap 1.1 ties for first on Artificial Analysis Decision Index with 384K context — ArtificialAnlys · 2026-10-23
- PolyAI unveils Dialog-RSN-1: audio-native voice model answering in under 300ms — matthen2 · 2026-10-23