WhatsApp pages accidentally exposed hidden AI chatbot instructions
About 80 help pages on WhatsApp's site carried concealed directions for its AI support bot, including a ban on discussing the GDPR. The company blames a bug and says the text has been taken down.
By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .
This story is no longer being updated. It reflects reporting up to .
What happened
Around 80 pages on WhatsApp's website exposed instructions meant for its AI support chatbot, including a line reading WHAT NOT TO SAY. One instruction told the bot not to recommend data protection authorities, and another told it not to discuss the GDPR even when users asked. WhatsApp says a bug caused the text to be exposed and has removed it.
Why it matters
The company says the leak was a bug and that the material is now gone.
Timeline
- Oct 19, 12:00 PM ET: Around 80 WhatsApp help pages were found to contain hidden instructions for its chatbot, including WHAT NOT TO SAY.
- Oct 19, 12:00 PM ET: Meta said a bug caused the instructions to be exposed.
- Oct 20, 3:01 AM ET: WhatsApp said the exposed text had been removed from its help pages.
Coverage
- WhatsApp help pages told its AI not to mention privacy regulators (notebookcheck.net)
- WhatsApp Help Pages Leaked Hidden Instructions for a Meta Chatbot (wired.com)
More from Technology & AI
- Eight states sue OpenAI, alleging misleading AI safety claims
- OpenAI announces solution to Navier-Stokes math problem
- Datadog bets AI agents can handle testing of AI-written code
- Anthropic and OpenAI sketch wealth-sharing plans before IPOs
- Mamdani administration proposes AI customer service rules
- Florida engineers study how AI behaves in high-risk situations