Instructions intended for WhatsApp’s support chatbot appeared in the underlying code of around 80 help and FAQ pages, according to an analysis by privacy company Redact. The hidden text included guidance on legal questions, accessibility-related messages and features that were not yet generally available.
Meta said a bug had caused the chatbot instructions to be displayed in the pages’ code. The company removed them after WIRED contacted Meta about the exposure.
Chatbot directions were embedded in help pages
Redact found 100 blocks of text written to guide a large language model. The instructions appeared in the code of pages on WhatsApp’s website, alongside content meant to answer users’ questions about the messaging app.
The blocks appeared under the heading “HelpContentGalaxyLLMInclude.” While some directions pointed users toward official support channels, the text also revealed how the chatbot was meant to handle certain subjects. WIRED said archived versions of the pages still showed the exposed material in their underlying code.
One example appeared on a page about deleting a WhatsApp account. Among other directions, the chatbot was told: “Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority” about the issue.
Legal questions came with strict limits
Instructions on a page about reporting and blocking told the chatbot how to respond when a user mentioned legal action, lawsuits, lawyers, attorneys, courts, or litigation. The directions said not to cite or interpret WhatsApp’s Terms of Service, Privacy Policy, or other legal documents, and not to make statements about the company’s liability.
They also said: “Do not provide legal opinions, defenses, or interpretations on behalf of WhatsApp.” Separate instructions told the chatbot not to cite or interpret data protection and privacy regulations, including GDPR, LGPD, and CCPA, even when a user raised those regulations directly.
WhatsApp spokesperson Joshua Breckman said the chatbot was intended to provide product support rather than legal advice. He said the company gives users transparent information about their data rights and how they can take action.
Meta said a bug exposed instructions meant for its AI
Breckman said the text was intended for the chatbot, not for users, and was designed to help WhatsApp Support AI provide accurate and relevant responses to people who choose to message it. Meta also said it trains its support chatbots to be as helpful as possible.
The exposure raised a distinction between the advice shown to users and the directions shaping the chatbot’s replies. Dan Saltman, CEO and founder of Redact.dev, said many of the instructions were straightforward and directed people to official support channels. He also said the text did not appear intended for public exposure.
It was unclear whether third-party chatbots could access the instructions or how long the text had appeared in the pages’ code. Meta said the bug had been fixed, and the instructions were no longer visible in the live pages’ code.
Other rules addressed accessibility and unreleased features
Not all the exposed guidance concerned law or privacy. One instruction explained how to treat a user saying they were deaf or hard of hearing, describing that information as background rather than the support request itself.
Another direction told the chatbot not to mention specific upcoming community features that were not yet generally available. Together, the examples showed that the hidden text covered both how the bot should interpret a user’s message and what information it should avoid offering.
The disclosure offered a view into how chatbot instructions can sit alongside public help content in a website’s code. Meta characterized the exposure as a bug and said the material was intended to shape responses from its own support chatbot.