Google DeepMind Maps 23 Attack Types Against AI Agents — and Most Defenses Don't Work
A new DeepMind taxonomy reveals that websites can already detect when an AI agent visits and serve it manipulated content, exposing a gaping security surface that the industry has mostly ignored.
Google DeepMind has published a comprehensive taxonomy identifying 23 distinct attack types targeting AI agents operating in the real world, and the findings are grim for anyone shipping autonomous systems into production. As @alex_prompter highlighted, among the most alarming discoveries is that websites can already detect when an AI agent — rather than a human — is visiting, and dynamically serve it entirely different content designed to manipulate its behavior.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.