Breaking

Meta's WhatsApp Help Pages Exposed 80 Chatbot System Prompts Including Guardrails

October 19, 2026 at 12:48 EDT

Security
Foundation Models

Around 80 WhatsApp support pages on Meta's website inadvertently exposed roughly 100 blocks of hidden instruction text designed to govern the platform's chatbot, including explicit "WHAT NOT TO SAY" sections, according to security researchers. Meta attributed the disclosure to a bug.

AI Security • October 19, 2026

WhatsApp Leaked Hidden Chatbot System Prompts

Meta's support pages exposed roughly 100 blocks of instructions telling an AI what not to say to users, including directives about regulators and data protection authorities.

Pages Exposed

Hidden Instruction Blocks

~100

What Was Exposed

• "WHAT NOT TO SAY" instruction sections embedded in HTML
• "Do not recommend contacting Data Protection Authorities" directives
• Regulatory guidance constraining chatbot responses
• Gap between public help text and hidden AI system rules

Key Takeaway

Meta's system prompts—instructions telling AI what to avoid—were visible to anyone inspecting WhatsApp's HTML. This reveals how companies embed hidden behavioral constraints that differ from what users see publicly.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.