A Comparative Evaluation of AI Agent Security Guardrails

arXiv cs.AI / 4/29/2026

💬 OpinionIdeas & Deep AnalysisModels & Research

Key Points

  • The paper compares DKnownAI Guard with AWS Bedrock Guardrails, Azure Content Safety, and Lakera Guard for securing AI agents against real-world risk scenarios.
  • Using human annotations as ground truth, it evaluates guardrails on two risk types: attacks targeting the agent (e.g., instruction override, indirect injection, tool abuse) and requests aimed at generating harmful content.
  • Results show DKnownAI Guard achieves the top recall rate of 96.5%, indicating it most effectively catches relevant risks.
  • DKnownAI Guard also leads overall on true negative rate (TNR) at 90.4%, suggesting fewer false alarms relative to competitors.
  • The study concludes DKnownAI Guard delivers the best combined performance across the tested AI agent security guardrails.

Abstract

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.

A Comparative Evaluation of AI Agent Security Guardrails | AI Navigate