Meta has accidentally exposed hidden instructions designed for its AI-powered WhatsApp support chatbot, revealing how the company guides the system when responding to users. The instructions appeared in the underlying code of nearly 80 WhatsApp help and FAQ pages, according to a WIRED investigation based on findings from privacy firm Redact.
Some of the instructions included sections labeled ‘WHAT NOT TO SAY.’ They told the chatbot to avoid giving legal advice and, in certain situations, to avoid discussing privacy regulations, legal documents, or Meta's potential liability.
The exposed material was intended to guide an AI system rather than provide information directly to users. However, its appearance in publicly accessible webpage code offered an unusual look at the rules shaping an automated customer support experience.
What did the Leaked WhatsApp AI Instructions Reveal?
The exposed text included guidance for handling questions about account deletion, privacy concerns, and possible legal action. In one example, the chatbot was instructed not to recommend contacting a data protection authority or regulator when users asked about deleting their accounts.
Other instructions told the AI not to quote or interpret WhatsApp's terms of service, privacy policy, or other legal documents. They also directed it to avoid discussing regulations such as the European Union's General Data Protection Regulation (GDPR), Brazil's LGPD and California's CCPA.
The instructions also covered less sensitive situations. For example, the chatbot was told not to mention specific upcoming community features that weren't yet generally available. These rules appear to be part of the company's effort to keep automated support responses within defined boundaries. However, the language around regulators and privacy laws has raised questions about how the chatbot handles users seeking information about their rights.
Also Read: Meta’s Algorithms Under Scrutiny: What Is MeitY Asking?
Meta Said a Bug Exposed the Hidden Text
WhatsApp spokesperson Joshua Breckman said a bug caused the text to appear, and the company has since fixed it. According to the company, the instructions were designed to help its support chatbot provide accurate and relevant answers to people who choose to contact WhatsApp Support using AI. Meta said the guidance was intended for the chatbot, not for users viewing the help pages.
It remains unclear how long the instructions were publicly accessible or whether third-party AI systems accessed them before Meta removed them.