AI traffic flowing between applications, agents, models, and external services carries a class of threats that traditional security tools don't inspect — prompt injections, sensitive data leakage, malicious content in AI-generated output, and agent-specific attacks such as memory poisoning and tool misuse. Prisma® AIRS™ AI Runtime Security detects and blocks these threats in real time as traffic flows through your AI applications and agents.

You enable AI Runtime Security through two approaches, and you can use one or both. Both modes share the same policy framework and security intelligence, so you configure policies once and they apply consistently across your entire deployment.

Protection spans five areas: prompt injection and jailbreak prevention (AI Model Protection), memory poisoning, tool misuse, and hallucination attack blocking (AI Agent Protection), sensitive data leakage prevention (AI Data Protection), malicious URL and malware detection in AI-generated content (AI App Protection), and harmful and toxic content moderation (AI Safety). Some capabilities — including contextual grounding for RAG applications and in-line data redaction — are currently available on API Intercept only.

Network Intercept deploys in three models to match your infrastructure: on public cloud (AWS, Azure, GCP), in Kubernetes environments using Hyperscale Security Fabric (HSF), or in private cloud and on-premises environments using the microperimeter model. You can combine deployment models across your environment.