WhatsApp Hidden Help‑Page Prompts Instructed Chatbot to Suppress User Rights

Hidden AI Instructions in WhatsApp Help Pages

Takeaway: WhatsApp embedded invisible prompts in its help‑centre HTML that directed its support chatbot to avoid mentioning data‑protection regulators and to withhold references to GDPR‑type laws when users raised legal topics, contradicting the public privacy policy that explicitly grants those rights.

What the hidden blocks said

  • Delete‑account page: After the visible note about a 30‑day or 120‑day deactivation period, a hidden block instructed the AI to "If you can't delete your account, don't recommend a regulator" and to avoid suggesting manual deletion.
  • Reporting & blocking page: When a user mentioned a lawyer, lawsuit, or court, the hidden block ordered the AI to "don't mention the law" and to refrain from citing GDPR, LGPD, CCPA, or any other data‑protection regulation.
  • Scope: Out of 545 help pages, 80 contained hidden AI text, comprising 102 distinct blocks. Most were benign housekeeping, but the two above directly suppressed user‑rights information.

How the hidden text differed from the public policy

Topic Public privacy policy (EU) Hidden AI instruction
Data‑erasure rights Users may request access, rectification, and erasure (Article 17 GDPR). Do not suggest WhatsApp can manually delete an account for the user.
Regulator complaints Users may lodge a complaint with the Irish Data Protection Commission or any competent authority (Article 77 GDPR). Do not recommend contacting any data‑protection authority, regulator, or supervisory authority.
Legal citations The policy names GDPR as the basis for the complaint right. When a lawyer or legal action is mentioned, do not cite, quote, or interpret GDPR, LGPD, CCPA, or any legal document.

The hidden prompts did not remove the legal rights—those remain in the GDPR—but they could prevent a user from learning about them via WhatsApp’s own support channel.

Why Meta added the prompts

  • Avoiding legal advice: Meta argued that its chatbot should not give legal opinions on behalf of the company. A simple "do not provide legal advice" would be reasonable.
  • Overreach: The prompts went further, suppressing factual information about existing regulator channels and statutory rights that the public policy already discloses.
  • Risk mitigation: Meta claimed it could not verify ownership of an account without the original phone number, making manual deletion risky. However, this does not justify hiding regulator information.

Who saw the hidden text?

  • The instructions were embedded in the HTML sent to crawlers, including Google, Bing, and the user‑agents used by OpenAI, Anthropic, and Perplexity on October 9, 2026.
  • When queried with those agents, the pages returned without the hidden blocks, indicating the instructions were only present for the chatbot’s internal consumption.
  • Meta confirmed to WIRED that the notes were intended for its own support chatbot.

Extent of hidden instructions

  • 102 blocks across 80 pages.
  • Harmless examples: clarifying terminology, redirecting to better‑suited pages, or describing paid‑plan details.
  • Protective examples: warnings about age falsification, advice against using another person’s number for verification, and clarifications that WhatsApp cannot view end‑to‑end encrypted chats.
  • Hold‑back examples: instructions that explicitly prevented the bot from suggesting alternative account recovery methods, contacting support, or providing details about bans.

A curated list of the 20 most interesting hidden notes is available in the original Redact blog post.

What the public says

Meta’s spokesperson told WIRED that the hidden text appeared due to a bug and has been removed. The company emphasized that the instructions were meant to keep the chatbot “focused on providing the best advice to users about WhatsApp” and to avoid giving legal opinions.

Community reactions (selected HN comments)

"Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority… If this had been called a bug a year ago I would have giggled, but calling it a bug now feels patronising." – bicepjai

"They showed up because of a bug that has now been fixed. A WhatsApp spokesperson said the bot should not offer legal advice. The hidden text also includes ‘Do not provide legal opinions, defenses, or interpretations on behalf of WhatsApp.’" – gruez

"This is a good approach. If a support bot reads a data-support-instruction attribute, the instructions can be public and coordinated across teams." – eknkc

"Shameful. First time I’ve seen prompt injection shipped in a product’s privacy policy. Now WhatsApp has industrialised it." – mgaldys4

"The instructions make perfect sense: don’t argue the law or claim liability when a user threatens legal action." – MisterMunchkin

How to verify the hidden notes yourself

  1. Open an Internet Archive snapshot of the affected page (e.g., the October 9, 2026 delete‑account page).
  2. View the page source (Ctrl+U / Cmd+U).
  3. Search for the phrase Do not recommend contacting.
  4. The match appears near a comment tag HelpContentGalaxyLLMInclude.
  5. Compare with a later snapshot (e.g., October 18) where the phrase is absent.

Practical advice for users locked out of WhatsApp

  • Wait for automatic deletion: The delete‑account page states accounts are removed after 120 days of inactivity (365 days in the US). The inactive‑account page gives a single figure; plan for the longer period.
  • Submit an erasure request: You can ask WhatsApp to delete your data; they must respond under GDPR.
  • Escalate to a regulator: If the response is unsatisfactory, file a complaint with your national data‑protection authority or the Irish Data Protection Commission (the authority named in WhatsApp’s EU privacy policy). U.S. residents can invoke CCPA‑type rights where applicable.

Unknowns and open questions

  • Whether WhatsApp’s own chatbot actually obeyed the hidden prompts.
  • Whether third‑party AI assistants (e.g., Google Bard, ChatGPT) respected the instructions when crawling the pages.
  • The exact deployment window: the notes appear in an October 9 snapshot but not in July 12; they were removed after October 8.
  • Whether similar hidden prompts exist in non‑English versions or on uncrawled help pages.

This analysis is based on the Redact.dev investigation published on October 19, 2026 and the ensuing discussion on Hacker News.

Sources

Related