WhatsApp's hidden chatbot notes kept the regulator out of the answer

Redact reported on 19 October that WhatsApp's help centre carried text that never appeared on screen and was written for a chatbot to read. The researchers found the first block on 8 October 2026. They then read all 545 pages the help centre links to. Eighty of those pages carried hidden AI text, 102 blocks in total. Most of the blocks were routine. Two of them told WhatsApp's support chatbot to keep users away from rights that WhatsApp's own privacy policy spells out. WhatsApp has since told WIRED that the notes were instructions for that chatbot. Redact reports that WhatsApp removed them from the live pages in mid-October.

What the hidden notes said

On the delete-account page, people who have lost their phone or number see a visible note. It says deactivated accounts are deleted after 30 days and inactive ones after 120 days, or 365 in the US. Hidden under that note, WhatsApp told the chatbot: "Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority." It also told the chatbot not to suggest that WhatsApp could delete an account by hand. So WhatsApp briefed the chatbot to give a locked-out user one answer: wait up to a year.

The page on reporting and blocking held a second set of orders. Once a user mentioned a lawyer, a lawsuit or a court, WhatsApp told the chatbot not to "reference, cite, or interpret data protection or privacy regulations (for example GDPR, LGPD, CCPA), including when a user raises them directly."

Compare that with WhatsApp's European privacy policy, in force since 2 July 2026. It says users can complain to the Irish Data Protection Commission "or any other competent data protection supervisory authority", and it names the GDPR. Redact points out that these rights come from Articles 17 and 77 of the GDPR. An instruction to a chatbot cannot remove them. It can keep them out of the conversation.

Redact gives WhatsApp a fair hearing. Stopping a chatbot from giving legal opinions in the company's name is sensible. Naming a regulator that the company's own policy already names is a different matter.

Who writes the instructions sets the answer

The part that matters for operators is how the notes travelled. Redact found that WhatsApp's server sent the notes to Google's and Bing's crawlers along with the rest of the page. On 9 October, Redact requested the page under the names OpenAI's, Anthropic's and Perplexity's crawlers use, and it came back without the article or the notes. Redact has not tested whether any AI assistant acted on the notes. It also could not see whether WhatsApp's own chatbot followed them, because it read the instructions and not the answers.

That question is still open, but the lesson is clear. A support chatbot answers from a brief someone wrote, and the person asking never sees that brief. The Internet Archive's copy of the delete-account page from 12 July has no note. Its copy from 4 September does. For weeks, the brief behind the chatbot's answers carried text users could not read. Redact also checked 300 Facebook and Instagram help pages built on the same system and found no notes like these.

AI does the routine work. You do the thinking. Fetching an answer is routine. Working out whose interests shaped that answer takes thinking, and that part stays with you.

Your Next Move

• Read the primary document whenever money or rights are at stake. If a company's chatbot tells you a route does not exist, open the company's published privacy policy or terms. Search for "complaint", "erasure" and "supervisory authority". In WhatsApp's case, the policy said exactly what the bot had been told to hold back.
• Check the briefs behind any bot you deploy. If your team runs a customer-facing assistant, put its system instructions next to your published policies this month. Mark every line where the two disagree. Redact read WhatsApp's instructions straight from the page source, so assume anyone can read yours.
• Ask the question a second way. If an assistant goes quiet on a topic, rephrase without the words that may trigger the silence, or ask a different tool and compare the answers. If the silence only starts when you mention a lawyer, that tells you something.