WhatsApp embedded 102 hidden text blocks across 80 of its 545 help centre pages — text rendered invisible to human readers but served to AI systems including its own support chatbot. Two blocks stood out: one instructed the AI never to recommend contacting a data protection authority to users who couldn't delete their accounts, and another told it to stop referencing GDPR, LGPD, or CCPA the moment a user mentioned a lawyer or legal action. The instructions directly contradicted WhatsApp's own published privacy policy for Europe, effective July 2, 2026, which explicitly names the Irish Data Protection Commission and cites the GDPR as the legal basis for user rights including data erasure and the right to complain. The hidden notes didn't just omit this information — they actively ordered the AI to withhold it. The account deletion case is structurally revealing. WhatsApp only allows account deletion from inside the app on the user's primary device. Lose the phone or number and you're locked out. The visible help page tells you to wait — up to 120 days, or 365 in the US. The hidden text told the AI that waiting was "the ONLY correct answer" and to never suggest the user contact a regulator, even though Article 17 and Article 77 of the GDPR guarantee the right to erasure and the right to lodge a complaint. The legal-trigger block was narrower but more aggressive. Once a user mentioned a lawyer, lawsuit, or court, the AI was told to avoid referencing, citing, or interpreting GDPR, LGPD, or CCPA — even when the user themselves raised the law. This is the opposite of what the privacy policy does: the policy names the GDPR and tells users it protects them. WhatsApp's server sent these hidden blocks not just to its own chatbot but to Google's and Bing's crawlers. When researchers tested under crawler names used by OpenAI, Anthropic, and Perplexity on October 9, the pages came back without the notes. Facebook and Instagram help centres, built on the same system, showed no equivalent hidden instructions across 300 pages checked. The researchers acknowledge limits: they cannot confirm whether WhatsApp's chatbot actually followed the instructions, whether outside AI assistants acted on them, or exactly when the blocks were placed (earliest confirmed: September 4, 2026; absent from July 12 copies). The notes were removed by mid-October 2026, after the discovery. The reasonable defence — that a company shouldn't let chatbots give legal advice — doesn't survive contact with the specifics. Telling someone a regulator exists is not legal advice. Naming a law that protects them is not legal advice. WhatsApp's own privacy policy does both. What the hidden notes did was create a second, invisible layer of policy that systematically narrowed the information available to users at the exact moment they needed it most.
