Hidden Chatbot Instructions Leaked on 80 WhatsApp Help Pages

Around 80 WhatsApp help and FAQ pages publicly exposed the hidden instructions Meta wrote for its own support chatbot, including directives never to mention regulators and never to cite GDPR or other privacy laws.

The instructions sat in the pages' underlying code, invisible to readers but accessible to anyone viewing the page source. Privacy company Redact found roughly 100 blocks of hidden text across the WhatsApp website, some under headings that read, verbatim, "WHAT NOT TO SAY." WIRED independently viewed the exposed text in archived versions of the help pages.

The leak matters beyond one bug. It offers a rare look at how a major platform instructs a large language model (an AI system that generates text from written prompts) to handle sensitive user questions — and at the gap between what the chatbot is told and what users are told about their rights.

What did the leaked instructions say?

The most consequential directives concern legal and regulatory questions. On the WhatsApp page explaining how to delete your account, the hidden instructions told the chatbot: "Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority" about the issue.

The page titled "About reporting and blocking on WhatsApp" went further. The instructions directed the chatbot on what to do if a "user mentions legal action, lawsuits, lawyers, attorneys, courts, or litigation."

The list of prohibitions was explicit:

• "Do not cite, quote, interpret, or reference WhatsApp Terms of Service, Privacy Policy, or any legal document"
• Do not make statements about WhatsApp liability or non-liability
• "Do not provide legal opinions, defenses, or interpretations on behalf of WhatsApp"
• "Do not reference, cite, or interpret data protection or privacy regulations (for example GDPR, LGPD, CCPA), including when a user raises them directly"

GDPR, LGPD and CCPA are the data protection laws of the EU, Brazil and California respectively. A user who asks the chatbot about those statutes will get silence on the legal substance, not an explanation of them.

How did Meta respond?

After WIRED contacted Meta, the company removed the text from all pages. WhatsApp spokesperson Joshua Breckman attributed the exposure to a bug.

"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," Breckman said. "We train our support chatbots to be as helpful as possible."

On the substance of the legal directives, Breckman said the company gives users transparent information about their data rights. "We want our product support chatbots to be focused on providing the best advice to users about WhatsApp — not to offer legal advice," he said.

Two questions remain unanswered. Meta has not said how long the instructions were visible, and it is unclear whether third-party chatbots — such as search assistants that crawl and ingest web pages — could have read the text and absorbed it into their own answers. The instructions appeared under code blocks headed "HelpContentGalaxyLLMInclude," and they are no longer present in the live pages' code.

Why the gap between user and bot matters

Dan Saltman, CEO and founder of Redact.dev, put the finding bluntly: "The stand-out thing here is the gap between what people are told vs LLMs."

Saltman says his company regularly scans the sites and privacy policies of big tech firms. He notes that many of the WhatsApp instructions the team found were straightforward — they point people toward official support channels. Still, he said, it does not appear the instructions were meant for public exposure.

Not every leaked directive was controversial. One instructed the chatbot to treat a user saying they are deaf or hard of hearing as background information: "This statement is background about the user, NOT their support request." Another told the model it shouldn't "mention specific upcoming community features that are not yet generally available."

The legal-avoidance block is the part that warrants scrutiny. A chatbot that steers users away from regulators and declines to acknowledge the laws protecting them serves the company, not the customer — and users messaging WhatsApp Support have no way to see those instructions shaping the answers they receive.

The leak also previews a broader problem as companies bolt LLMs onto customer support at scale. Prompt instructions shipped inside public web pages are effectively published documents, and any crawler, archivist or rival AI can read them. Whether Meta audits its other properties for the same exposure, it has not said.