Whatsapp's hidden chatbot guidelines exposed in leaked help pages

A recent leak revealed hidden instructions for WhatsApp's chatbot embedded in the app's help pages, according to an analysis by privacy firm Redact. The exposed text, found in the code of 80 help and FAQ pages, included guidance for a large language model, specifically one used by Meta. The instructions, such as 'Do not recommend contacting any Data Protection Authority,' were not meant for public exposure. Dan Saltman, CEO of Redact.dev, highlighted the discrepancy between user guidance and chatbot instructions. After WIRED contacted Meta, the text was removed, with WhatsApp spokesperson Joshua Breckman attributing the exposure to a bug. The instructions, labeled 'HelpContentGalaxyLLMInclude,' aimed to ensure chatbots provided accurate support without offering legal advice. Breckman emphasized that WhatsApp's chatbots are trained to assist users, not provide legal counsel. The leaked content also included guidelines on handling users who are deaf or hard of hearing and avoiding mentions of upcoming features.