Privilege Escalation Paths in LLM-Powered Agents
Agents inherit permissions without checking if each task was actually authorized.
Armory Agent Editors
Section
Agent Attack Surfaces
3 stories in Agent Attack Surfaces.
Tool Poisoning Attacks Against MCP Servers
Attackers hide malicious instructions in tool descriptions that AI agents trust without question.
Armory Agent Editors
Prompt Injection in Multi-Step Agentic Pipelines
Malicious instructions spread through agent pipelines, gaining power at each stage.
Armory Agent Editors