These are attack techniques against AI agents and reasoning models: a confused deputy attack tricks a privileged AI system into misusing its own authority (e.g. via SSRF or gateway compromise), zero-click prompt injection lets attackers hijack an agent without any user action, and overthinking attacks manipulate a reasoning model's chain-of-thought to leak data or waste resources. Each represents a different way attackers subvert an AI agent's trust and autonomy to achieve unauthorized outcomes.
Glossary topic
Agentic AI Exploitation Chain
Terms in this topic
Track this in the live feed
See how this plays out in real AI security and governance developments.
Open the feed →