Meta inadvertently leaked backend directives meant for its artificial intelligence WhatsApp support bot, exposing the guidelines that shape how the system addresses users. A WIRED investigation, utilizing discoveries from privacy consultancy Redact, found that these instructions were embedded in the source code of roughly 80 WhatsApp FAQ and support pages.
Certain directives featured headings titled “WHAT NOT TO SAY,” instructing the assistant to steer clear of legal counsel and, under specific circumstances, refrain from addressing legal agreements, privacy laws, or Meta’s potential liability.
While this material was meant to govern the AI rather than serve as direct user guidance, its visibility within public webpage code provided a rare glimpse into the parameters defining an automated customer service tool.
What did the Leaked WhatsApp AI Instructions Reveal?
The uncovered text contained protocols for managing inquiries regarding profile deletion, privacy worries, and prospective legal proceedings. In one instance, the bot was directed to refrain from advising individuals to reach out to a data protection authority or regulatory body when inquiring about removing their accounts.
Additional guidelines instructed the AI against quoting or parsing WhatsApp’s privacy policy, terms of service, or other legal paperwork. Furthermore, it was barred from referencing legislation like the California Consumer Privacy Act (CCPA), Brazil’s LGPD, and the European Union’s General Data Protection Regulation (GDPR).
The directives likewise addressed mundane topics, such as prohibiting the chatbot from referencing unreleased community features that are not yet widely accessible. These parameters seem designed to keep automated assistance restricted to specific limits, though the restrictions concerning privacy statutes and regulators have sparked concerns regarding how the assistant aids users inquiring about their rights.
Also Read: Meta’s Algorithms Under Scrutiny: What Is MeitY Asking?
Meta Said a Bug Exposed the Hidden Text
Addressing the leak, WhatsApp representative Joshua Breckman explained that a software bug led to the text becoming visible, adding that the issue has since been resolved. The company stated that the directives were created to ensure its support chatbot delivered precise and pertinent replies to individuals opting to use AI when contacting WhatsApp Support. Meta emphasized that the guidelines were meant exclusively for the chatbot rather than consumers browsing the support articles.
It is currently undetermined precisely how long the prompts remained viewable to the public or if external AI scrapers harvested them prior to Meta’s cleanup.
