ActiveFence AI Security Benchmark Report
1 NoamBarr 1 8/20/2025, 2:33:03 PM activefence.com ↗
Comments (1)
NoamBarr · 6h ago
This benchmark compares six AI security detection models on accuracy, precision, recall, F1, and false-positive rate (FPR): ActiveFence’s AI Safety & Security model; three open-source baselines (deepset DeBERTa-v3-base-injection, Meta Llama-Guard-2-8B, ProtectAI DeBERTa-v3-base-prompt-injection-v2); and two commercial APIs (Amazon Bedrock Guardrails—PROMPT_ATTACK—and Azure Content Safety—ShieldPrompt).ectAI, Bedrock guardrails and Azure.