WhatsApp's help pages carried hidden notes telling an AI not to recommend a privacy regulator
Redact's write-up shows the visible and hidden versions of each page side by side, so you can judge the wording yourself.
Redact, a data-deletion company, says it found text on WhatsApp's help pages that never appeared on screen and was meant for an AI to read. Of 545 help-centre pages it saved, 80 carried 102 hidden blocks. Redact says most were harmless, but two stood out. On the account-deletion page, a hidden note told the AI not to recommend contacting any data protection authority or regulator for users locked out of their accounts. On the reporting-and-blocking page, a note told it to stay away from legal topics, including privacy law, once a user mentioned a lawyer or lawsuit.
Redact says the notes were gone by October 18. It reports that WhatsApp told WIRED the notes were instructions for its support chatbot that showed up on the pages because of a bug now fixed, and that WhatsApp wants its support chatbots focused on WhatsApp advice, "not to offer legal advice."
-
General Data Protection Regulation
After exploring WhatsApp's Hidden Notes Told AI Not to Recommend a Regulator, the law the notes steered around is the General Data Protection Regulation.
Covers the EU's privacy regulation, the rights it gives people over their data and the role of national data protection authorities.
Wikipedia · Swim · 30 min
0 -
LLM01:2025 Prompt Injection
After exploring WhatsApp's Hidden Notes Told AI Not to Recommend a Regulator, why text people can't see still steers an AI is explained in LLM01:2025 Prompt Injection.
OWASP explains how inputs can alter a language model's behavior, and notes that such text need not be visible to humans as long as the model reads it.
OWASP Gen AI Security Project · Wade · 5 min
0