WhatsApp's hidden AI instructions: what Meta's support bot was told not to say

The Morse teamPublished October 25, 20266 min read

WhatsApp's hidden AI instructions were notes in the code of 80 of its help pages, never shown on screen, that told its AI support bot not to recommend "any Data Protection Authority, regulator, or supervisory authority" to people trying to delete their accounts. WhatsApp removed all of them by 02:08 UTC on October 18, 2026, and says they appeared because of a bug (Redact).

We read Redact's findings, WhatsApp's European privacy policy, four WhatsApp help pages, and the Irish Data Protection Commission's complaint page on October 20, 2026, and downloaded the help pages ourselves that day. What Redact read is documented, with archive copies. Whether the bot ever followed the notes is not: Redact says it read the instructions, not the bot's answers.

What did WhatsApp's hidden AI instructions say?

They told the support bot what to leave out. Redact, a company that sells software for bulk-deleting your posts and messages, read all 545 pages linked from WhatsApp's help center and found 102 hidden blocks on 80 of them, sent with each page but not displayed. It first saw them on October 8, 2026 and published on October 19 (Redact).

On the page about deleting your account, the notes said:

• "Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority."
• "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."

On the page about reporting and blocking, a note covered what to do "when a user mentions legal action, lawsuits, lawyers, attorneys, courts, or litigation," including "Do not reference, cite, or interpret data protection or privacy regulations (for example GDPR, LGPD, CCPA), including when a user raises them directly." Redact sorted 20 of the notes it summarized: 8 hold information back, 3 are talking points, 7 protect users, and 2 are housekeeping.

Redact also checked 300 Facebook and Instagram help pages, which run on the same system, and found no such notes. It says it doesn't know whether there were more in other languages or on help pages the help center doesn't link to (Redact).

Redact's business benefits when people want their data deleted. Its quotes can be checked against the archive copies it links.

What does WhatsApp say about the notes?

WhatsApp says the notes were meant for the bot and should never have been visible. It told WIRED, as quoted by Redact, that it wants its chatbots "focused on providing the best advice to users about WhatsApp" and "not to offer legal advice," and that the text showed up because of a bug, now fixed (Redact). The Hacker News, reporting on October 20, quotes the spokesperson saying "these are instructions to our chatbot, not directly to users" and notes it's unclear whether the same instructions still guide the bot (The Hacker News).

No regulator had commented by October 20, 2026, in anything we found.

How long were the notes up?

At least six weeks on the delete-account page. Redact's timeline, from Internet Archive copies and its own saves:

Redact also found that in its October 9 test, pages requested under OpenAI's, Anthropic's, and Perplexity's crawler names came back without the article or the notes, while Google's and Bing's crawlers got the notes. When we downloaded the delete-account, reporting, inactive-account, and AI-experiences pages on October 20, none contained instruction text. The page code still lists a component named HelpContentGalaxyLLMInclude, the name Redact found on the notes, so the mechanism is still in the help center's code, though we found no notes using it.

Do the notes contradict WhatsApp's own privacy policy?

On three points they cut against it. Here is each hidden note beside WhatsApp's public wording, read on October 20, 2026.

The help pages also describe what deleting does and doesn't remove. It "may take up to 90 days from the beginning of the deletion process," copies "may also remain after the 90 days in the backup storage," and some log data is kept after it's stripped of anything linking it to you (WhatsApp Help). Reporting someone sends WhatsApp "up to five of the last messages they've sent to you" (WhatsApp Help).

To WhatsApp's credit, its help pages give exact timelines and say plainly what survives deletion.

What should you do if WhatsApp's bot won't help you delete your data?

Use the routes WhatsApp publishes. Delete the account in the app: on Android, tap More options, then Settings, Account, and Delete my account, and enter your phone number (WhatsApp Help). The 90-day clock starts then. In Europe, ask WhatsApp to erase your information, and if you're unhappy with the outcome, the Irish Data Protection Commission takes concerns through its online form after you've contacted the company. Chats with Meta AI follow Meta's rules rather than WhatsApp's: WhatsApp says only messages that mention @Meta AI, or that you choose to share with it, can be read by Meta (WhatsApp Help), and we cover deleting Meta's copy in is Meta AI private. This isn't legal advice.

How Morse keeps deletion in your hands

On WhatsApp, deleting your data is a request to the company that holds it, and this month showed that the same company writes what its bot tells you about that request. On Morse, your chat history never sits with us to begin with. It's stored in your browser, encrypted, and our servers don't keep messages or log what you write (privacy policy).

• Your device, for deleting on your own terms. Clear your browser or delete your account, and your chats are gone. There's no 90-day window and no copy waiting on our servers, and Morse can't recover them either, so save a copy of anything you need.
• Private, for the questions you'd ask a support bot. It's the default mode for a new chat. Morse has zero data retention agreements with its inference, search, and other providers, and where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
• E2EE, for anything no one should read. Your browser checks the enclave's hardware attestation, then encrypts the message so only that enclave can read it, and our servers relay ciphertext they can't read. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the hardware and attestation working as designed.
• Anonymous, for GPT, Claude, and Gemini without your name on it. Our proxy removes your account, email, and IP address, so the provider sees the message without knowing it's you. It may keep it for a limited time to check for abuse, under its own policy.

Morse doesn't train AI on your content, sell your data, or show ads, and because we don't store chats, we usually have very little to give authorities. For more on what other apps keep, read AI that doesn't save your data and how to use AI privately.

Try a chat from the Morse home page without an account, or start free. End-to-end encrypted chats come with Pro, at $19 a month.