Hidden AI chatbot instructions exposed on WhatsApp pages

Hidden instructions for an AI support chatbot appeared in the underlying code of nearly 80 WhatsApp webpages, including guidance on handling legal and privacy questions. Meta said a bug exposed the text and that it had been fixed. The instructions were removed from live pages after Meta was alerted, according to reports. Privacy firm Redact found nearly 100 blocks of hidden text while examining WhatsApp help pages. One account put the total at 102. The text, in code blocks labelled “HelpContentGalaxyLLMInclude,” was not part of the pages’ visible help content but could be viewed in their source code. Some instructions addressed what the chatbot should do when users raised legal issues. They included guidance not to cite or interpret WhatsApp’s Terms of Service, Privacy Policy or legal documents, and not to discuss data-protection or privacy regulations. Another set, on an account-deletion page, reportedly told the chatbot not to recommend contacting a data-protection authority or regulator. Meta spokesperson Joshua Breckman said the exposure resulted from a bug that had been fixed. He said the material was intended to guide AI responses to users who choose to contact WhatsApp...