WhatsApp help pages leaked hidden Meta chatbot instructions — brand-safety lesson for AI deployments

Context

Around 80 WhatsApp help pages exposed instructions meant for Meta's chatbot, including a "WHAT NOT TO SAY" section, which Meta attributes to a bug. The incident shows how internal system prompts can end up public, with reputational consequences for any brand whose customer-facing AI is built on similar guardrails. Marketing and CX leaders deploying chatbots should review how their own prompt and content governance is handled.