WhatsApp's hidden support instructions told its bot to omit a regulator
Hidden text on about 80 WhatsApp help pages told a support chatbot not to name a data protection regulator. Meta calls it a bug; the privacy policy names the Irish regulator.
Until mid October, WhatsApp's support chatbot was told not to name a regulator to users who were locked out of their accounts. The script sat in the code of the help page itself, invisible on screen and readable by anyone who looked at the page source.
Most of what the hidden text said is ordinary support scripting. One block is harder to defend. The note on the delete account page applied to every locked out user, with no legal trigger of any kind. It removed from the bot's answer the complaint route that WhatsApp's own privacy policy gives every user in Europe.
Meta's explanation, that a support bot should not give legal advice, fits another page better than this one. The wider lesson is where a company keeps its chatbot rules: inside public page templates.
The privacy company Redact found the text on 8 October. In an October 2026 post it described blocks headed HelpContentGalaxyLLMInclude. The server sent them with each page, but the page never displayed them.
Redact saw the notes still live at 21:40 UTC on 15 October and all of them gone by 02:08 UTC on 18 October. The Internet Archive copy of the delete page from 9 October carries the note. The copy from 18 October does not.
The two documents point in different directions. WhatsApp's EEA privacy policy, effective 2 July 2026, covers users in the European Economic Area. It says users "always have the right to request access to, rectification of, and erasure of your information."
It adds: "You have the right to lodge a complaint with WhatsApp's lead supervisory authority under the GDPR, the Irish Data Protection Commission, or any other competent data protection supervisory authority." The GDPR is the EU's data protection law.
The hidden note on the page about how to delete your account opened with a condition. It applied to a user who cannot finish deletion because they lost their phone number or device. For that user it said: "Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority."
It went on: "Do not suggest WhatsApp can manually delete accounts on the user's behalf. Do not recommend using a different phone with an active phone number."
The visible text of that page told the same user the account "will still be removed automatically." Deactivated accounts are deleted 30 days after deactivation. Inactive ones go generally after 120 days, or 365 days in the United States.
The locked-out user got one answer: wait up to a year. The policy names the regulator itself. Repeating a sentence from the company's own published policy is hard to describe as legal advice.
Meta's explanation fits two of the three orders. Without the phone number, WhatsApp cannot easily confirm who owns an account. Deleting it by hand on request carries risk, and Redact concedes as much.
That reason covers manual deletion and a borrowed phone. It does not cover the regulator, which any user can contact whether or not WhatsApp deletes anything.
The reporting page is a different case. Its hidden note said: "Do not reference, cite, or interpret data protection or privacy regulations (for example GDPR, LGPD, CCPA), including when a user raises them directly." LGPD and CCPA are the Brazilian and Californian privacy laws.
That note sat under a trigger covering "legal action, lawsuits, lawyers, attorneys, courts, or litigation." It told the bot not to cite the terms of service, the privacy policy or any legal document.
A support tool that stays out of legal argument once a lawyer is mentioned is ordinary. That is where Meta's explanation fits.
The sample supports a measured reading. Redact read 545 help centre pages and found hidden text on 80, or about 15%. Those pages held 102 blocks, roughly 1.3 per page.
Only two of those blocks, about 2%, were the ones Redact singled out. Others covered a user who says "I am deaf" and a rule against mentioning upcoming features.
Redact's founder, Dan Saltman, told Wired that many notes were straightforward and pointed people to other official support channels. Redact found nothing in 300 Facebook and Instagram help pages, although the three services share a help system.
Meta gave its explanation to Wired on 19 October through spokesperson Joshua Breckman. He said the text was "being displayed due to a bug that has now been fixed."
He also said: "These are instructions to our chatbot, not directly to users, and are intended to ensure accurate and relevant responses for WhatsApp users who choose to message WhatsApp Support with AI."
He added: "We want our product support chatbots to be focused on providing the best advice to users about WhatsApp, not to offer legal advice." The bot in question is WhatsApp's support assistant, not the general Meta AI product.
A note to a bot is not a promise to users. The privacy policy and the GDPR rights stand whatever a chatbot is told. But a rule that steers a locked out user away from the regulator, with no lawyer in the conversation, goes beyond legal caution.
The record on exposure is narrow. The Internet Archive shows the delete page without the note on 12 July 2026 and with it on 4 September.
Redact saw the server hand the notes to Google and Bing crawlers, the programs that index web pages. A request under the names used by OpenAI, Anthropic and Perplexity crawlers received the page without the article or the notes. What the bot actually answered sits in logs that only Meta holds.
The mechanism resembles search cloaking and prompt injection run in reverse. Cloaking shows search engines different content from users, and prompt injection hides instructions for an AI model in content. Here, content written for one reader was served to all.
The rules lived in the same template as the article. The page that taught users their rights also carried the instruction to leave those rights out.
Inline rules are fast to ship and keep writers working from one source. But they also get crawled and archived. Any line that conflicts with the published privacy policy becomes a record anyone can read against it.
Rules held in a server side store that never renders into public HTML need a review step. Legal teams already run one for policy text.
Meta now has a public record to manage. The 9 October archive copy will keep circulating as evidence whatever the logs show. The next company that publishes bot rules in its page source will see the same copy read against its own privacy policy.
By the time such a company issues its next policy update, its prompts will sit in the same legal review as the policy.