A security oversight involving WhatsApp’s help center resulted in the public exposure of internal documentation intended for a Meta chatbot. The leaked data, which spanned roughly 80 individual web pages, contained specific operational guidelines and behavioral constraints for the AI, including a section explicitly titled “WHAT NOT TO SAY.” These instructions are typically used by developers to define the boundaries and safety protocols for AI interactions.

Meta has officially acknowledged the incident, characterizing the exposure as the result of a technical bug rather than a malicious breach. The company stated that it is taking steps to rectify the issue and prevent similar occurrences in the future. While the pages were accessible via search engine indexing, it remains unclear how long the information was publicly available before being identified and addressed. The incident highlights the ongoing challenges tech companies face in managing the security of internal AI development documentation as they integrate generative AI features into widely used messaging platforms.