AI Agent Security Self-Assessment
How much harm could a manipulated AI agent do?
Interactive example
Where the real security boundary sits
Select each layer to see the risks and controls at that boundary. The key idea: the model cannot be relied on to separate instructions from data, so authorization must be enforced outside it.
Authorization boundary
This is the control that actually limits harm: tools check the requesting user’s rights, validate parameters and require human approval for consequential actions.
Controls at this boundary
- Tools enforce the user’s permissions and tenant
- Narrow, task-specific tools
- Human approval with exact parameters
- Write tools disabled after untrusted input
Illustrative model only. No model inference or agent execution happens on this page.
What this assessment is — and is not
It is a structured, self-reported review of an AI system’s design. It does not test for prompt injection, trace agent execution or scan anything, so it reports no verified exposures. Self-reported gaps and hypothetical attack categories are shown separately.Assess your AI system
Learn more
MITRE ATT&CK® and ATT&CK® are registered trademarks of The MITRE Corporation. ATT&CK data © 2026 The MITRE Corporation. This work is reproduced and distributed with the permission of The MITRE Corporation (ATT&CK v20.1). MITRE ATLAS™ data (v2026.10) © MITRE, used under the Apache License 2.0. Securitain is not affiliated with or endorsed by MITRE. Explanations, scenarios and control mappings on this page are Securitain’s own analysis. Sources and licensing