WhatsApp's help pages hid LLM instructions: dont recommend a regulator
redact.devds
WhatsApp's help documentation contained hidden instructions for language models that advised against recommending regulators, according to analysis of the platform's publicly accessible pages. The discovery raises questions about how AI training data is being sourced from corporate websites.
Why it mattersIf major platforms are embedding hidden directives in public pages intended for AI systems, it suggests deliberate attempts to shape model behavior during training—a practice that could affect how AI assistants respond to questions about regulation and compliance.
Read at redact.dev