WhatsApp’s help center briefly exposed about 80 pages of internal instructions written for its AI chatbot. The pages included a section labeled “WHAT NOT TO SAY.” Meta says a bug caused the exposure.
The leaked material offered a rare look at how the company directs its chatbot’s behavior. Instructions covered tone, topics to avoid, and handling of sensitive questions. The pages were publicly accessible on WhatsApp’s website.
Meta confirmed the issue and attributed it to a technical bug. The company said the pages were not meant for public view. It did not specify how long the material remained accessible.
The disclosure highlights growing scrutiny of AI systems and the rules that govern them. Chatbot instructions often shape how assistants respond to users. When exposed, they can reveal editorial choices and guardrails.
Such instructions typically include guidance on avoiding harmful or misleading statements. They may also set boundaries around medical, legal, or political topics. The WhatsApp pages followed that pattern.
Meta has faced repeated questions about transparency in its AI products. Publishing or leaking internal prompts can invite outside analysis. Regulators and researchers increasingly request visibility into how chatbots are trained and constrained.
The company says the bug has been addressed. It has not said whether the exposed pages will be restored in another form. Users are unlikely to notice any change in the chatbot’s behavior.
