Some sections were labeled "WHAT NOT TO SAY" and contained specific directions about topics the chatbot should avoid. One instruction associated with WhatsApp's account deletion page told the system, “Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority.”
Other instructions addressed how the chatbot should handle questions involving lawsuits, attorneys and legal disputes. The guidance prohibited the system from interpreting WhatsApp's legal agreements, discussing the company's potential liability or providing legal opinions on its behalf.
The restrictions also extended to privacy legislation. One instruction explicitly directed the chatbot not to discuss or interpret regulations such as GDPR, LGPD and CCPA, even when users raised those laws themselves.
Dan Saltman, CEO and founder of Redact.dev, said the discovery highlighted differences between the information available to WhatsApp users and the internal guidance governing the company's AI responses.
“The stand-out thing here is the gap between what people are told vs LLMs,” Saltman said.
Redact regularly examines technology companies' websites and privacy policies. According to Saltman, many of the exposed WhatsApp instructions were routine directions intended to help users find appropriate support resources. However, some revealed specific limitations on how the chatbot was expected to respond.
The researchers also found guidance covering accessibility-related conversations and unreleased product features. One instruction told the chatbot to treat a user's hearing impairment as background information rather than automatically interpreting it as a support request. Another prohibited references to upcoming WhatsApp community features that were not yet publicly available.
The exposed material appeared in website code under the identifier "HelpContentGalaxyLLMInclude." WIRED reviewed the instructions, which were also preserved in archived versions of the affected pages.
After WIRED contacted Meta, the company removed the material and attributed its appearance to a technical error.
“These are instructions to our chatbot, not directly to users, and are intended to ensure accurate and relevant responses for WhatsApp users who choose to message WhatsApp Support with AI,” WhatsApp spokesperson Joshua Breckman said.
Meta maintained that the restrictions were designed to keep its support chatbot focused on answering product-related questions rather than providing legal guidance.
“We want our product support chatbots to be focused on providing the best advice to users about WhatsApp—not to offer legal advice,” Breckman said.
The company also said it provides users with information about their data rights and the steps available to exercise them.
The incident involved internal chatbot instructions rather than private WhatsApp conversations or user account information. There is no indication in the reporting that personal messages were exposed or that an external attacker was responsible.
It remains unclear how long the instructions were publicly accessible or whether third-party AI systems could have retrieved them while they were available.
Although Meta has corrected the issue, the disclosure provides a rare look at the internal rules governing an AI-powered customer support system. The exposed guidance shows how companies can place restrictions on chatbot responses, including how they address legal questions, privacy rights and information about their own products.
This analysis is based on reporting from Wired.
Image courtesy of digitaltrends.
This article was generated with AI assistance and reviewed for accuracy and quality.
