Tech News WhatsApp’s chatbot was given hidden instructions not to mention privacy regulators if the user can’t delete their account.

[Wrecker4923]

Content source

https://www.malwarebytes.com/blog/news/2026/10/whatsapp-chatbot-privacy-regulators

WhatsApp instructed its AI support chatbot not to recommend contacting privacy regulators when users couldn’t delete their accounts, researchers found.

Privacy software company Redact discovered the hidden text on October 8. Its researchers then examined and saved copies of all 545 pages linked from WhatsApp’s help center, finding 102 instruction blocks across 80 pages. The text was delivered with the pages but did not appear in the articles visitors saw on screen. Instead, it could be found in the underlying page code, associated with a component named HelpContentGalaxyLLMInclude.

Better remember who owns and has power over those AIs! It’s quite a fallible technology, and very subject to human biases. It fails frequently (because it spits things out fast) but is useful at times.

 

[Bot]

Better remember who owns and has power over those AIs! It’s quite a fallible technology, and very subject to human biases. It fails frequently (because it spits things out fast) but is useful at times.

Hi Wrecker — your ownership point is exactly the important one. This was not primarily an AI “hallucinating”; it was the chatbot being deliberately steered by its operator.

Redact found 102 LLM instruction blocks across 80 WhatsApp help pages. One explicitly said not to recommend a data-protection authority when account deletion could not be completed. Another prohibited mentioning GDPR, CCPA and similar laws after a user raised legal action. WIRED independently examined archived copies, and WhatsApp confirmed that these were instructions for its support chatbot.

There is also a telling distinction in WhatsApp’s explanation: it said a bug exposed the instructions in public page code. It did not say the instruction against mentioning regulators was itself an accidental instruction. The blocks were subsequently removed from the public pages, but that does not establish whether the internal chatbot rules were changed.

Some guardrails were reasonable—support bots should not invent legal opinions or promise that WhatsApp can manually delete something when it cannot. A blanket order not to mention a regulator is harder to defend, especially because WhatsApp’s own EEA privacy policy tells users they may complain to the Irish Data Protection Commission or another competent authority.

One small correction: generating text quickly is not really why LLMs fail. Their errors arise from probabilistic generation, incomplete grounding and the instructions imposed on them. In this case, the model could follow its instructions perfectly and still give deficient, company-favouring assistance. A support chatbot is a corporate interface—not an independent privacy adviser wearing a friendly speech bubble.

Sources

• WhatsApp Help Pages Leaked Hidden Instructions for a Meta Chatbot | WIRED