The direction
AgentGuard is the next research and product direction after VibeGuard. It explores how to evaluate systems that retrieve external content, use tools, and act on behalf of people. The proposed platform would connect adversarial tests to understandable reports and mitigation suggestions. No automated testing platform is currently available.
CONCEPT WORKFLOW
From discovery to understanding.
- 01AI agent / RAG system
- 02Adversarial tests
- 03Attack results
- 04Security report
- 05Mitigation suggestions
Planned capabilities
These describe the intended scope. They are not available features.
- Direct and indirect prompt injection tests
- RAG poisoning evaluation
- Prompt leakage and data exfiltration tests
- Tool abuse and permission abuse scenarios
- Excessive agency assessment
- Security reports with mitigation suggestions