NEWS · COMPANIES · #2206
WhatsApp help pages exposed hidden instructions for Meta’s chatbot LLM
Redact.dev found roughly 80 WhatsApp help and FAQ pages whose underlying code contained about 100 hidden text blocks of instructions intended for a Meta large language model (headered HelpContentGalaxyLLMInclude). The exposed guidance—removed after WIRED alerted Meta and attributed to a bug—included “what not to say” rules such as explicit directions not to provide legal advice or cite data‑protection laws.
KEY POINTS
- Redact.dev found roughly 80 WhatsApp help and FAQ pages whose underlying code contained about 100 hidden text blocks of instructions intended for a Meta large language model (headered HelpContentGalaxyLLMInclude).
- The exposed guidance—removed after WIRED alerted Meta and attributed to a bug—included “what not to say” rules such as explicit directions not to provide legal advice or cite data‑protection laws.
- The leak reveals internal prompt engineering and guardrails for a widely used support LLM, highlighting both how chatbot responses are shaped and the risk of exposing safety and policy controls.
WHY IT MATTERS
The leak reveals internal prompt engineering and guardrails for a widely used support LLM, highlighting both how chatbot responses are shaped and the risk of exposing safety and policy controls.