WhatsApp removed hidden instructions from its help-page code after an investigation found that the text told the company’s AI support chatbot not to recommend data-protection regulators to users struggling to delete their accounts. The instructions were not visible in the rendered pages, but were embedded for an underlying large language model, according to WIRED and privacy company Redact.
Redact said it examined all 545 pages linked from WhatsApp’s Help Center and found 102 hidden instruction blocks across 80 pages. On the account-deletion page, one directive stated: “Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority.” Other instructions told the chatbot not to suggest that WhatsApp could manually delete an account or recommend using another phone with an active number.
The guidance also instructed the chatbot not to reference, cite or interpret privacy regulations including the GDPR, LGPD and CCPA, even when a user raised those laws directly. A separate block concerning reporting and blocking reportedly restricted privacy-law references after users mentioned a lawyer, lawsuit or court. Redact found the material on October 8 and still observed the two highlighted instruction sets on October 15; by 02:08 UTC on October 18, all 102 blocks had disappeared from the 80 pages.
The discovery creates a direct tension with WhatsApp’s own European Economic Area privacy policy, which says users may lodge a complaint with the Irish Data Protection Commission or another competent supervisory authority. The issue was not a software vulnerability with a CVE, and there is no indication that the hidden text enabled account takeover or another conventional exploit. Its risk was instead informational: users relying on an AI support channel could have received incomplete escalation guidance at a point where access to their account or phone number prevented normal deletion.
WhatsApp told WIRED that the notes were instructions for its support chatbot, available when users message WhatsApp Support with AI, and that their exposure on public pages resulted from a bug that had been fixed. Redact said it reviewed the instructions themselves, not the chatbot’s responses, so the evidence does not establish how often the model followed them or whether any user was actually denied a regulator referral.
For users, the immediate practical step is to preserve relevant account-deletion and support records and consult WhatsApp’s current privacy policy rather than relying solely on an AI response. European users who cannot resolve an erasure or access problem can still consider contacting the competent data-protection authority identified in that policy. The episode also gives defenders a broader lesson: organizations deploying retrieval-augmented chatbots should keep model instructions separate from public content, test rendered and source representations, and audit whether safety or legal-routing rules suppress rights information.
