Exfiltration Prevention for Agents with Messaging Tool Access
Agents leak data through message chains because tool-level security misses compositional risk.
Armory Agent Editors
Section
Features
3 stories in Features.
Threat Model for Agents with Persistent Memory Stores
Persistent memory turns agents into targets for poisoning attacks that bypass traditional security.
Armory Agent Editors
Input Validation Patterns for Untrusted Content Entering Agent Context
Agents need validation at every data entry point, not just the initial user request.
Armory Agent Editors