Two blocks out of 102
Start with the proportion, because it cuts in WhatsApp's favour before it cuts against it. Most of the 102 blocks Redact found were housekeeping, per Redact's own summary: clearing up words users mix up, pointing a question at a better page, describing paid plans. A few protected users, including one telling the AI never to suggest entering a false age. If the story were 102 blocks of corporate instruction, there would be no story. Companies instruct their support bots. That is what support bots are.
Two blocks were different in kind, not degree.
The first sat under the visible note on the delete-account page. That page matters because of who lands on it: per the visible text Redact reproduces, WhatsApp only deletes an account from inside the app, on the main phone, with the number still working. Lose the phone or the number and the normal route is closed. The visible text offers waiting: deletion 30 days after deactivation, or after 120 days of inactivity, 365 in the US. The hidden text repeated those periods and then closed the remaining doors. Do not recommend a regulator. Do not suggest WhatsApp can delete the account by hand. Do not recommend using a different phone with an active number. Had the support AI followed that text, a locked-out user who asked it would have been given one answer — wait up to a year — and would not have heard from it that any other route existed. Whether it followed the text is untested. Redact did not test it.
The second block sat on the page about reporting and blocking, and its trigger is the revealing part. It applied once a user mentioned legal action, a lawsuit, a lawyer or a court. From that point the AI was not to cite WhatsApp's Terms of Service or Privacy Policy, not to discuss WhatsApp's liability, and not to reference, cite or interpret data protection or privacy regulations, including when the user raised them directly. In other words, the moment a conversation turned legal, the laws a user might rely on were to leave the conversation too.
The policy on the same website says the opposite
None of this required leaked documents to check. WhatsApp's privacy policy for the European region is a public document the company is required to publish, effective 2 July 2026, and it is more open than the hidden notes on both of the points that matter.
On complaints, the policy gives users the right to lodge one with the Irish Data Protection Commission or any other competent supervisory authority. The hidden note told the AI not to recommend contacting any of them. On erasure, the policy says users always have the right to request access to, rectification of, and erasure of their information. The hidden note told the AI not to suggest WhatsApp could delete an account on a user's behalf. Redact is careful on that second pair, and the care is worth keeping: the note did not ban mentioning erasure requests, it banned suggesting a manual deletion. The gap there is narrower. On the regulator, there is no narrower reading available. The two texts point in opposite directions.
The rights themselves do not come from WhatsApp. In the EU they sit in Article 17 and Article 77 of the GDPR. An instruction to a chatbot cannot remove them. What it could do is affect whether a user learns about them through the support channel, the place they went for help.
A bug explains the exposure. It does not explain the text
In a statement to WIRED, as relayed by Redact and The Hacker News, Meta spokesperson Joshua Breckman said the text consisted of internal instructions for WhatsApp's own support chatbot, exposed by a bug that has since been fixed. He added that the instructions were meant to ensure accurate and relevant responses for people messaging WhatsApp Support with AI, and that the company wants its support bots to give advice about WhatsApp rather than legal advice. WIRED's original page was not directly accessible to us, so the statement reaches this article through those two reports, which relay the same statement rather than independently confirming it.
Take the strongest version of that defence, because part of it holds. A company is entitled to stop its chatbot giving legal opinions in the company's name, and directing someone to a qualified legal professional is a reasonable instruction. Had the reporting-and-blocking block stopped at legal advice, this would be a story about a formatting bug.
It did not stop there. Naming a law that protects a user is not, in our view, legal advice, and neither is telling that user a regulator exists; that is this site's editorial judgement, not a legal determination. WhatsApp's own privacy policy does both, in public, on the same website. So the question the bug defence leaves open is not how the text became visible, which it answers, but why a no-legal-advice rule was written as a no-regulator, no-GDPR rule, applied specifically once a user mentioned a lawyer. There is also a second open question the removal does not settle: the instructions are gone from the pages, but nothing published so far says whether the chatbot itself still works from the same guidance internally.
And one limit should stay attached to every version of this story, including this one. Redact read instructions, not answers. It did not test whether WhatsApp's chatbot actually withheld the regulator from anyone. The documented harm is a standing instruction, not a recorded refusal.
What a locked-out user can actually do
Redact reports that WhatsApp's help pages describe different waiting periods for account deletion, including 30 days after deactivation, 120 days of inactivity and a 365-day period in the US. The pages reportedly disagree in places, so these figures should not be treated as one universal rule. Check the current instructions that apply to your account and jurisdiction.
A user can also send WhatsApp an erasure request directly; WhatsApp's published European privacy policy states that right without a waiting period. If the answer is unsatisfactory, that policy names the further route: a complaint to the user's own national data protection authority, or to the Irish Data Protection Commission, the authority the policy names. California and a growing number of US states give residents a deletion right as well, on their own terms rather than the European policy's.
These routes existed in WhatsApp's published documentation while the hidden instructions were present. Redact found instructions restricting what the support AI should recommend or cite; it did not establish that the chatbot actually withheld these options from users. The distinction matters especially for account deletion: the quoted instruction prohibited recommending manual deletion by WhatsApp, not mentioning an erasure request as such.
Sources
- [1] Redact, "WhatsApp's Hidden Notes Told AI Not to Recommend a Regulator", redact.dev, 19 October 2026 — primary source: the full instruction wording, the 545 / 80 / 102 counts, the crawler tests, the archive timeline, the policy comparison and the list of what remains unknownRead source
- [2] The Hacker News, "WhatsApp Hid Instructions in Help Pages Telling Its AI Chatbot Not to Recommend Privacy Regulators", 20 October 2026 — write-up carrying the same quoted blocks and the GDPR Article 17 / 77 references; it relays the same WIRED statement as Redact and is not independent confirmation of that statementRead source
- [3] Notebookcheck, "WhatsApp help pages told its AI not to mention privacy regulators", 20 October 2026 — independently checked the account-deletion sentence word for word against an Internet Archive copy from 9 October 2026; that check covered one sentence, not the full set of pagesRead source
- [4] Note on sourcing: WIRED's original reporting is the source of WhatsApp spokesperson Joshua Breckman's quotes, as reported by Redact and The Hacker News. WIRED's page itself could not be fetched directly, so those quotes are attributed through the two reports above rather than to a direct readRead source