# AI Agent Security & Threat Detection Tools: evaluation worksheet

Source: https://dutygraph.com/directory/ai-governance/categories/agent-security/
Editorial date: 2026-09-07

## Scope

- Organization / team:
- Task and expected output:
- Human owner:
- Product and version:
- Evaluation date / environment:
- Reviewer:

## Questions

### 1. Which prompt-injection and tool-abuse surfaces are tested, and which are excluded?

- Observation (demonstrated / described / unknown):
- Evidence reference:
- Limitation or follow-up:

### 2. Does the product prevent the action or notify someone after it occurred?

- Observation (demonstrated / described / unknown):
- Evidence reference:
- Limitation or follow-up:

### 3. How are benign unusual requests separated from attacks, and how can false positives be reviewed?

- Observation (demonstrated / described / unknown):
- Evidence reference:
- Limitation or follow-up:

### 4. What evidence is available for response without exposing unnecessary sensitive content?

- Observation (demonstrated / described / unknown):
- Evidence reference:
- Limitation or follow-up:

## Evidence checklist

- [ ] A scoped adversarial test report
- [ ] A prevented or detected event with its action timeline
- [ ] False-positive review and incident-response procedures

## Boundary to check

No single demonstration establishes complete protection. Combine security testing with least privilege, data controls and task boundaries. A security alert does not replace a business decision about whether the agent should have been doing that work.

## Decision

- Fit for the scoped task:
- Unresolved gaps:
- Next action, owner and date:

This is a planning worksheet, not an endorsement, access approval or compliance certification. Keep confidential evaluation notes in your organization's approved storage.