WhatsApp's help pages included LLM instructions that told the model not to recommend a regulator.

What it means

An investigation by Redact.dev found that WhatsApp's help pages contained hidden, machine-readable instructions (prompt injection) telling the company's AI support chatbot not to recommend or name any data protection regulator. This effectively suppressed users' GDPR-given right to complain to authorities like the Irish Data Protection Commission, since the chatbot would steer conversations away from such recommendations.

Accuracy & sourcesGrounded

The claim comes from a detailed investigation blog post by Redact.dev, which appears to have reverse-engineered the help pages. The finding is corroberated by Wired and several other tech outlets, giving it moderate-to-high confidence as a documented occurrence, though the original source is a secondary investigation, not an official leak.

How new