On October 8, 2026, Dan Saltman, founder of the data deletion service Redact, discovered a hidden paragraph within the source code of WhatsApp’s "delete account" help page. The text was not intended for human readers, but functioned as a system instruction for an AI support bot. One of the directives explicitly stated: do not recommend contacting any data protection authority, regulator, or supervisory body.1

To determine the scope of these hidden directives, the team cataloged 545 pages linked through the help center. They uncovered 102 hidden blocks across 80 different pages.1

Key Findings
  • System instructions distinguish between the visible text for users and the hidden guidance provided to the AI bot.
  • The bot is subject to specific prohibitions when dealing with users who have lost access to their phone or account number.
  • Instructions explicitly forbid the bot from discussing legal regulations or data protection rights even when prompted by the user.

What are the invisible AI instructions in WhatsApp help pages?

WhatsApp help pages are delivered to browsers as HTML files, a portion of which is rendered for the user. Embedded within these files were blocks labeled "HelpContentGalaxyLLMInclude," where LLM refers to the large language model technology powering AI chatbots. These blocks remain invisible to standard users but are accessible to anyone inspecting the page source code. The WhatsApp support bot reads these help pages to inform its responses, meaning these embedded notes function as system prompts that constrain the bot's behavior. While many of the 102 blocks identified by Redact are benign—such as clarifying terminology or directing queries to relevant pages—others serve as guardrails. For instance, the bot is instructed not to suggest that minors enter false birth years or that users delete groups where they have been defrauded, instead advising them that chat histories may serve as evidence.1

How does the hidden guidance affect users unable to delete accounts?

WhatsApp accounts can typically only be deleted from within the application on the primary phone while the number is active. Users who lose their device or their phone number are often unable to perform this action. While the public-facing help page notes that inactive accounts are generally deleted after 120 days—or 365 days in the United States—the hidden instructions impose strict limitations on the bot. The AI is forbidden from recommending that users contact a regulator, from suggesting that WhatsApp can manually delete the account, or from advising the user to attempt the process from another device. For users who have lost their number, the system defines waiting as the "only" correct answer.1

A second category of restrictions appears on pages regarding complaints and blocking. When a user mentions legal terms such as lawsuits, attorneys, or courts, the bot is instructed to avoid mentioning WhatsApp’s terms of service or privacy policies, and to refuse to comment on corporate liability. Furthermore, even if a user explicitly asks about data protection laws—such as Europe's GDPR, Brazil's LGPD, or California's CCPA—the bot is directed not to discuss them. The instructions often conclude by directing the user to consult an attorney.12

Notes and Sources

  1. Saltman, D. (October 19, 2026). "WhatsApp's Hidden Notes Told AI Not to Recommend a Regulator". Redact. redact.dev
  2. Khandelwal, S. (October 20, 2026). "WhatsApp Hid Instructions in Help Pages Telling Its AI Chatbot Not to Recommend Privacy Regulators". The Hacker News. thehackernews.com
  3. Burgess, M. (October 19, 2026). "WhatsApp Help Pages Leaked Hidden Instructions for a Meta Chatbot". WIRED. wired.com