WhatsApp Leaked Hidden Chatbot Instructions
WhatsApp help pages inadvertently exposed hidden instructions for a Meta chatbot due to a bug.
"Oops! WhatsApp accidentally showed its chatbot's secret rulebook. Turns out, even AI has 'what not to say' lists. Big tech, big oops."
Around 80 WhatsApp help and FAQ pages on its website exposed internal instructions designed for a Meta chatbot, including sections titled “WHAT NOT TO SAY.” Meta attributed the exposure to a bug, which has since been fixed. The instructions, intended for a large language model, were not meant for public display and appeared in the webpages’ code, according to findings from privacy company Redact.
The exposed text, which WIRED viewed in archived versions, included directions such as not to recommend contacting data protection authorities, regulators, or supervisory authorities. It also instructed the chatbot not to provide legal advice, interpret terms of service, or make statements about WhatsApp liability. A WhatsApp spokesperson stated these instructions were for their chatbot to ensure accurate responses and not to offer legal advice.
This incident provides a glimpse into how instructions for chatbots are created and managed, highlighting the distinction between information provided to users and internal guidance for AI systems. The text has been removed from the live pages’ code, and it remains unclear how long the instructions were accessible or if third-party chatbots could access them.
This incident underscores the importance of robust content management and security protocols for companies deploying AI. It highlights the potential for internal operational instructions to become public, impacting user trust and perception. Businesses must ensure a clear distinction between internal AI guidance and public-facing communications.
Relevant tools
Find the right AI tool for your business
Chat with Insta and get matched to the right tool in seconds.
Try Insta Tool Finder →