ML Security PapersA Comparative Evaluation of AI Agent Security Guardrails
Qi Li , Jiu Li , Pingtao Wei , Jianjun Xu , Xueyi Wei , Jiwei Shi , Xuan Zhang , Yanhui Yang , Xiaodong Hui , Peng Xu , Lingquan Zhou
Published on arXiv
2604.24826
Prompt Injection
OWASP LLM Top 10 — LLM01
Insecure Plugin Design
OWASP LLM Top 10 — LLM07
Key Finding
DKnownAI Guard achieves 96.5% recall and 90.4% true negative rate, outperforming AWS Bedrock Guardrails, Azure Content Safety, and Lakera Guard
This report presents a comparative evaluation of DKnownAI Guard in AI agent security scenarios, benchmarked against three competing products: AWS Bedrock Guardrails, Azure Content Safety, and Lakera Guard. Using human annotation as the ground truth, we assess each guardrail's ability to detect two categories of risks: threats to the agent itself (e.g., instruction override, indirect injection, tool abuse) and requests intended to elicit harmful content (e.g., hate speech, pornography, violence). Evaluation results demonstrate that DKnownAI Guard achieves the highest recall rate at 96.5\% and ranks first in true negative rate (TNR) at 90.4\%, delivering the best overall performance among all evaluated guardrails.
Key Contributions
- Comparative evaluation of 4 commercial AI agent guardrails (DKnownAI Guard, AWS Bedrock, Azure Content Safety, Lakera Guard) on 1,018 adversarial samples
- Focus on agent-specific attacks: instruction override, indirect injection, tool abuse, privacy leakage, and harmful content elicitation
- Human-annotated ground truth evaluation across 8 public adversarial datasets (ALERT, Salad-Data, Tensor-Trust, etc.)