Zscaler AI Runtime Protection (formerly SPLX) monitors and secures AI applications in production environments. The platform detects and blocks prompt-based threats including jailbreaks, prompt injections, and data leakage attempts in real-time. The solution applies input and output filtering to AI interactions, enforcing guardrails on both user prompts and model responses. Organizations can define custom policies using natural language to block specific topics, phrases, or behaviors based on their use case and compliance requirements. Detection thresholds can be adjusted to balance sensitivity and false positive rates. The platform provides detailed telemetry for security analysis, including logs of blocked prompts with timestamps, similarity scores, and trigger context. All monitoring and enforcement occurs with minimal latency impact on user experience. Integration capabilities include REST API connections, conversational platforms, and large language model frameworks. The platform supports both commercial and open-source AI models through pre-built connectors that require no coding to implement. The solution addresses AI security across the full lifecycle, from development testing to production runtime protection, with centralized visibility into AI risk posture and policy violations.