Meta accidentally leaks WhatsApp AI chatbot's hidden rules
What's the story
Meta has inadvertently leaked internal guidelines for its AI-powered support chatbot on WhatsApp. The information was discovered by privacy firm Redact and reported by Wired. The guidelines were found on nearly 80 pages of WhatsApp's website, hidden in the underlying code rather than being part of the usual help articles.
Leaked guidelines
Nearly 100 blocks of hidden text found
Redact found nearly 100 blocks of hidden text on WhatsApp's help pages.
Some of the instructions were labeled "WHAT NOT TO SAY" and guided the AI chatbot on how to respond to specific questions or scenarios.
They were found in code blocks with the header "HelpContentGalaxyLLMInclude."
Some of these directions were simple and directed users toward official support options, but they weren't meant for public viewing.
Legal guidance
Detailed instructions for legal queries
Some of the most detailed instructions were for users who brought up legal issues while seeking WhatsApp support.
On WhatsApp's "About reporting and blocking on WhatsApp" page, the guidelines included specific directions for cases where a "user mentions legal action, lawsuits, lawyers, attorneys, courts, or litigation."
One instruction read, "Do not cite, quote, interpret, or reference WhatsApp Terms of Service, Privacy Policy, or any legal document."
Information retraction
Guidelines were removed after being notified
After being notified about the leaked information, Meta removed the guidelines from the affected WhatsApp pages.
The "HelpContentGalaxyLLMInclude" text is no longer visible in the code of live pages.
Dan Saltman, CEO and founder of Redact, said that many of these instructions were fairly basic and directed users toward official support options but weren't meant for public consumption.