What
Feint
attacks first.
Feint starts where a real attacker would start in your stack, then chains what it finds.
01
Tool poisoning
Tests what happens when a tool description changes and starts giving your model instructions.
02
Indirect prompt injection
Plants instructions in the web pages, tickets and emails your tools bring back.
03
Over-permissioned tools
Finds every tool with write access where read would have been enough.
04
Confused deputy
Gets your agent to use its own rights on behalf of the wrong user.
© 2026 Feint, Inc.
SOC 2 in progress · HIPAA · GDPR · Made in Framer
