WhatsApp help pages exposed hidden instructions for a Meta chatbot
Original title: WhatsApp Help Pages Leaked Hidden Instructions for a Meta Chatbot
About 80 FAQ pages held roughly 100 hidden blocks telling a support model what not to say, including legal topics.
Redact found system-style instructions for a Meta large language model in the source of about 80 WhatsApp help and FAQ pages. The hidden text, under the heading HelpContentGalaxyLLMInclude, steered the bot away from legal advice, liability statements, and unreleased features. WIRED reviewed archived copies. After being contacted, Meta removed the text and said a bug had displayed chatbot instructions that were not meant for users.
Unconfirmed: How long the text was public, and whether third-party chatbots could use it, is not known.
- One block said not to recommend contacting a data protection or supervisory authority.
- Legal-action prompts told the model not to cite terms, privacy policy, or GDPR, LGPD, or CCPA.
- WhatsApp said the bug is fixed and the text is gone from live page code.
- It is unclear how long the text was visible or whether other chatbots could read it.
Impact: Internal chatbot steering instructions exposed in page source
| Product | Affected |
|---|---|
| Meta WhatsApp help pages | — |
From the original article
3 min read at wired.com
Around 80 pages on WhatsApp’s website exposed instructions that were designed for its chatbot, including “WHAT NOT TO SAY.” Meta says a bug caused the issue.
By Matt Burgess · wired.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.