Multiple outlets report that WhatsApp support pages briefly exposed hidden, internal instructions intended for a Meta AI chatbot. Wired says about 80 pages on WhatsApp’s website contained guidance for the chatbot, including sections such as “WHAT NOT TO SAY.” The reports describe the content as information not meant for public viewing.
Meta attributes the exposure to a technical issue, saying a bug led to the instructions appearing on the help pages. The outlets’ accounts focus on what the leaked material instructed the chatbot to avoid or follow, raising questions about how internal prompt or safety guidance can become accessible when surfaced through public web pages. Coverage also emphasizes the scope of the exposure and that Meta says it was unintentional.
At the time of reporting, outlets characterize the incident as a documentation or page-rendering problem rather than a deliberate disclosure, and they note that Meta’s response points to an underlying software defect.
