What happened
On Oct 29, Palo Alto Networks expanded its Prisma AIRS platform: its AI Gateway now applies bidirectional Model Context Protocol (MCP) guardrails (detecting hidden instructions in tool inputs, redacting sensitive output, enforcing access policies), adds Smart Routing to select models by task and track inference cost, integrates with Anthropic Claude Enterprise via inference hooks, and broadens AI software component scanning (models, skills, plugins, MCP servers, dependencies; 3M+ artifacts scanned).
Why it matters
This shifts AI security from inspecting prompts to governing agent activity — the gateway becomes a centralized enforcement point for agent-to-tool interactions and extends coverage to Claude Enterprise, directly relevant as agents gain operational authority. It is a major capability expansion from a top-tier security vendor in the fastest-growing AI-security category.
Applicability
Security leaders evaluating or using Prisma AIRS should test gateway policies against real tools and agent workflows; finance teams should measure routing savings per completed task, not per request.