WhatsApp's website briefly exposed the internal instructions that guide its AI-powered support chatbot. Privacy firm Redact found roughly 100 blocks of hidden text embedded in the code of about 80 help and FAQ pages. The text was not visible to users but was designed to steer Meta's large language model when answering support questions.
The instructions included explicit warnings about legal topics. For example, one page told the chatbot not to recommend contacting data protection authorities, while another said not to cite or interpret regulations like GDPR or CCPA. The guidance also covered how to handle users who mention lawsuits or say they are deaf or hard of hearing.
After WIRED contacted Meta, the company removed the hidden text from all pages. A WhatsApp spokesperson said the exposure was a bug and that the instructions were meant for the chatbot, not users. The company said the goal is to keep support chatbots focused on product advice rather than legal opinions. It remains unclear how long the text was visible or whether third-party chatbots could have accessed it.