WhatsApp Help Pages Exposed Hidden Chatbot Instructions, Including Bans on Legal Advice
A report from privacy firm Redact indicated that approximately 80 help and FAQ pages on WhatsApp's site contained concealed text intended for a large language model. This text, labeled "HelpContentGalaxyLLMInclude," included instructions to refrain from advising users to reach out to any Data Protection Authority and to avoid mentioning WhatsApp's Terms of Service or Privacy Policy in legal matters. Additionally, it advised against referencing data protection laws such as GDPR, LGPD, and CCPA. Dan Saltman, CEO of Redact, highlighted the inconsistency between user guidance and LLM directives. Following an inquiry from WIRED, Meta removed the text. WhatsApp's spokesperson, Joshua Breckman, explained that the issue stemmed from a bug and was meant for chatbot precision, not user support. The length of time this text was visible is uncertain.
Key facts
- Around 80 WhatsApp help and FAQ pages exposed hidden LLM instructions in their code
- The text appeared under the heading "HelpContentGalaxyLLMInclude"
- Redact, a privacy company, conducted the analysis and found roughly 100 blocks of hidden text
- One instruction barred recommending contact with any Data Protection Authority, regulator, or supervisory authority
- Instructions told the chatbot not to cite or interpret WhatsApp's Terms of Service or Privacy Policy during legal disputes
- GDPR, LGPD, and CCPA were named as regulations the chatbot should not reference
- Meta removed the text after WIRED inquired; spokesperson Joshua Breckman called it a bug that is now fixed
- Redact CEO Dan Saltman highlighted the gap between what users are told and what LLMs receive
Entities
Artists
- Dan Saltman
- Joshua Breckman
Institutions
- Meta
- Redact
- Redact.dev
- WIRED