ServiceNow-AI Unveils AprielGuard, a Safety Model for Agents

As LLMs evolve beyond simple text assistants into Agentic systems capable of tool use and reasoning, new threats are emerging that existing single-message-based safety classifiers struggle to handle.

AprielGuard is a 8B-parameter safety and security guardrail model designed to respond to these changes.

Key Features and Characteristics:

• Broad Detection Coverage: Detects 16 safety risk categories (toxicity, misinformation, illegal activity, etc.) as well as various adversarial attacks including prompt injection, jailbreaks, memory poisoning, and tool manipulation.
• Optimized for Agentic Workflows: Supports complex agentic environments including multi-turn conversations, long context, tool calls, and reasoning traces, not just simple prompts.
• Two Operating Modes: Provides both a Reasoning mode for cases requiring explainable classification and a Non-reasoning mode for low-latency production environments.

This model offers an integrated solution for managing the complex security threats arising in the modern LLM agent ecosystem.