AI Agent Security Guardrails Compared: DKnownAI Guard Leads Recall
A new report from arXiv compares DKnownAI Guard against AWS Bedrock Guardrails, Azure Content Safety, and Lakera Guard in detecting security risks for AI agents. Using human annotation as ground truth, the evaluation measures detection of threats like instruction override, indirect injection, and tool abuse, as well as harmful content requests such as hate speech, pornography, and violence. DKnownAI Guard achieved the highest recall at 96.5% and the best true negative rate at 90.4%, outperforming all competitors.
Key facts
- DKnownAI Guard achieved 96.5% recall rate.
- DKnownAI Guard ranked first in true negative rate at 90.4%.
- Benchmarked against AWS Bedrock Guardrails, Azure Content Safety, and Lakera Guard.
- Evaluation used human annotation as ground truth.
- Detected threats to agent: instruction override, indirect injection, tool abuse.
- Detected harmful content requests: hate speech, pornography, violence.
- Report published on arXiv.
- Title: A Comparative Evaluation of AI Agent Security Guardrails.
Entities
Institutions
- arXiv
- DKnownAI
- AWS
- Azure
- Lakera