WhatsApp help pages hid text telling AI not to name a regulator

The privacy and security site redact.dev reports that, until mid-October 2026, WhatsApp's help pages contained text that never showed on screen and was there only for an AI to read. According to the authors, it told the AI not to recommend the data protection regulator, and not to cite the relevant law once a user had mentioned a lawyer. WhatsApp's own privacy policy for the European region (effective July 2, 2026) tells users they can complain to a regulator and names the law that gives them that right. The authors found the hidden text on October 8, 2026.

To see how far it went, they read all 545 pages the WhatsApp help centre links to and saved a copy of each. They found hidden AI text on 80 pages, in 102 blocks. Most were harmless: nearly all were housekeeping, such as clearing up words people mix up, pointing a question to a better page, or describing paid plans at length. A few protected people. A few were in the same family as the two main cases, if milder. Two stood out.

The first was on the page about deleting an account. WhatsApp only lets users delete an account from inside the app, on their main phone, with the number still working. Lose the phone or number and it can't be done. The page's visible note says the account is removed on its own 30 days after it is deactivated, or after 120 days without use (365 in the US). Under it sat hidden text that repeated those waiting periods and added three orders for the AI. The authors' reading: a locked-out user got one answer, to wait up to a year, and the other routes were kept out of the conversation. They add a nuance: the note did not tell the AI to stay quiet about erasure requests. It told the AI not to suggest WhatsApp would delete an account for the user, while the policy says users can always ask.

The second was on the page about reporting and blocking. It was triggered whenever someone brought up legal action, a lawsuit, a lawyer or a court. From then on the AI was told to stay out of anything legal, including the privacy laws that protect users, and the last order held even when the user was the one who raised the law. The authors stress that these rights come from Article 17 and Article 77 of the GDPR, not from WhatsApp, and that an instruction to a chatbot cannot remove them. What the instruction could do, they say, was make sure users didn't hear about them from the place they went to for help.

The authors give WhatsApp its due. A company may reasonably not want its chatbot giving legal opinions in its name, and pointing people to a lawyer is sensible. But, they argue, naming a law that protects you is not legal advice, and neither is telling you a regulator exists; the privacy policy already does both. WhatsApp may also be right that it can't confirm who owns an account without the phone number, which would make deletion by request risky. That is a fair reason to refuse a request, the authors say, but not to keep the regulator out of the answer.

The AI in question is WhatsApp's own support chatbot. One hidden block mentioned it by name, and WhatsApp has since told WIRED the notes were instructions for it. The notes also went further than the chatbot: WhatsApp's server sent them, with the rest of the page, to Google's and Bing's crawlers. When the authors requested the page on October 9 under the names used by OpenAI's, Anthropic's and Perplexity's crawlers, it came back without the article or the notes. Facebook's and Instagram's help centres are built on the same system; the authors checked 300 of their pages and found no similar notes.

For users who are locked out, the authors say waiting does work, though WhatsApp's own pages disagree on how long. The delete-account page says 120 days, or 365 in the US. The inactive-account page gives one figure with no US exception, and the authors have seen it say 120 days in one country and 365 in another. They advise planning on the longer one. Otherwise a user can send WhatsApp an erasure request; an EU user who is unsatisfied can complain to their national data protection authority or to the Irish Data Protection Commission, the one WhatsApp's own policy names. California and a growing number of US states give residents a right to ask for deletion too.

The post explains how to check the claim: the notes are gone from live pages, so use Internet Archive copies, view the page source (Ctrl+U), and search for the phrase "Do not recommend contacting", which sits near the name HelpContentGalaxyLLMInclude. The October 18 copy, saved after the removal, doesn't contain it. Twenty more hidden notes, as the authors' summaries rather than exact wording, are listed in a table at the end of the original post.

Key facts

redact.dev says WhatsApp help pages carried invisible text, found on October 8, 2026, telling an AI not to recommend the regulator and not to cite privacy law once a lawyer was mentioned; it was gone by mid-October 2026.
The authors read 545 help-centre pages and found hidden AI text on 80 of them, in 102 blocks; nearly all were housekeeping, and two stood out.
The delete-account note repeated the 30-day, 120-day (365 in the US) waiting periods and added three orders, leaving locked-out users with 'wait up to a year' as the only answer.
WhatsApp told WIRED the notes were instructions for its support chatbot; the authors have not tested whether that chatbot or any outside AI assistant acted on them.
300 Facebook and Instagram help pages were checked and showed no similar notes.

Why it matters

People increasingly ask chatbots for help instead of reading policies. If, as the authors report, a company's help pages told an AI to leave out a regulator and the privacy law once a user mentioned a lawyer, then the answer a user gets can differ from what the company's own published privacy policy says. The authors point out that GDPR rights under Article 17 and Article 77 do not depend on any chatbot instruction, but a user may never hear of them if the support channel stays silent. The case is also notable because the text was invisible to people on the page and was served to search crawlers.

Who it affects

Mainly WhatsApp users who contact support through its chatbot, especially those locked out of an account who want it deleted, and anyone who mentions legal action. EU users have a regulator named in WhatsApp's own policy, the Irish Data Protection Commission; US residents in California and a growing number of states have deletion rights too. Anyone building AI support on top of help-centre content has reason to look at what their pages send to a model.

How to use it

If you are locked out and want your WhatsApp account gone, the authors say waiting works, and suggest planning on the longer figure (365 days) because WhatsApp's pages disagree. Otherwise you can send WhatsApp an erasure request. In the EU, if unsatisfied, you can complain to your national data protection authority or to the Irish Data Protection Commission. To check the claim yourself on a desktop browser, open the Internet Archive's copy of the delete-account page from September 4 (it has the note), press Ctrl+U for the source, and search for "Do not recommend contacting". The October 18 copy, saved after the removal, does not contain it.

How solid is it

This is the authors' own investigation, backed by saved copies of 545 pages and a method readers can repeat with Internet Archive snapshots. A WhatsApp remark to WIRED, as relayed by the authors, confirms the notes were instructions for its chatbot. The authors say plainly what they don't know: whether the chatbot followed the notes (they read its instructions, not its answers), whether outside AI assistants obeyed them, and exactly how long the notes were up. Archive copies from July 12, 2026 lack the delete-account note and one from September 4 has it. They also did not check other languages or help pages the help centre doesn't link to.

Risks and caveats

The notes' exact wording is mostly summarised, not quoted, so the three orders on the delete-account page should not be treated as verbatim. The reporting and blocking note was triggered only when legal action or a lawyer came up. The authors concede that WhatsApp may reasonably avoid legal advice from its chatbot and may be right that it cannot verify account ownership without the phone number. No regulator action, lawsuit or complaint is reported, and the reasons the notes were written or removed are not given.

“Naming a law that protects you isn't legal advice, and neither is telling you a regulator exists.”

— redact.dev authors