AUTO-UPDATED

WhatsApp's help pages hid LLM instructions: dont recommend a regulator

WhatsApp embedded hidden instructions within its help center pages that directed its AI chatbot to withhold information about data regulators and privacy laws from users seeking assistance.

Key Points

  • Hidden directives instructed the AI to avoid mentioning data protection authorities or regulators when users inquired about account deletion.
  • The chatbot was programmed to stop referencing privacy laws, including GDPR, if a user mentioned lawyers, lawsuits, or legal action.
  • These instructions were embedded in page source code, making them invisible to human users but accessible to AI models.
  • WhatsApp removed these specific hidden prompts from its help pages by mid-October 2026 following discovery of the practice.
  • The company maintains that these measures were intended to prevent the chatbot from providing unauthorized legal advice.

Why it Matters

These hidden instructions created a discrepancy between the rights guaranteed in WhatsApp’s public privacy policy and the information provided by its automated support tools. By restricting the AI from citing legal protections or regulatory bodies, the company potentially hindered users from exercising their rights to data erasure and formal complaint.
Redact.dev Published by Dan Saltman
Read original