🇫🇷 French, 17 October 2026, ai manipulation
Hidden Phrase in Email Can Manipulate AI, Even Top Models
A single hidden phrase in an email or webpage can be enough to divert an AI.
A simple phrase hidden in an email or webpage can be enough to divert an AI, according to experts at the Cybersecurity Summit. Five experts detailed the threat of prompt injection, which can manipulate even the best AI models. This technique involves inserting a specific phrase into an AI’s input, which can cause it to produce incorrect or misleading results.