As AI agents move from answering questions to writing code, using tools and executing workflows, they gain access to more of the business. Organisations need protection that keeps pace; otherwise either innovation slows down or control is lost.
Palo Alto Networks has announced new Prisma AIRS capabilities aimed at that balance, focused on three priorities:
- Secure frontier AI adoption with runtime protection integrated directly into Claude Enterprise.
- Protect the AI supply chain by extending scanning beyond models to agent skills and other components.
- Enforce control as agents act with new AI Gateway capabilities for tool security and smart routing.
Enterprise-grade security for Claude Enterprise
Tools like Claude are increasingly used for coding and business workflows. Prisma AIRS integrates with Claude Enterprise through inference hooks, bringing runtime protection to Claude Code, Claude.ai and Claude Cowork:
- Faster adoption with clear boundaries: runtime guardrails apply before model inference, and policy defines which interactions are allowed.
- Data protection without duplicate rules: existing enterprise DLP policies extend to keep PII, intellectual property and source code from being exposed, with no separate rule sets to maintain.
- Lower risk of AI breaches: prompt injections, jailbreaks and toxic inputs are blocked.
The AI supply chain beyond the model
Protecting interactions is only part of the picture. The models, agent skills, plugins, MCP servers and open-source dependencies behind AI apps must be secured too, because unscanned code can expose data and credentials.
With more than 3 million third-party artifacts pre-scanned according to Palo Alto Networks, Prisma AIRS AI Supply Chain Security scans open-source models, validates agent skills and blocks malicious code before deployment.
Real-time agent governance
Trusted components are the foundation, but how agents use models, tools and data in production must also be controlled. The Prisma AIRS AI Gateway centralises policy enforcement and usage oversight, reducing inconsistent controls and unpredictable costs. Since launch it has received more than 20 major updates, including support for decision models.
MCP guardrails: AI Gateway inspects MCP traffic in both directions, blocking concealed instructions in tool inputs, redacting sensitive data from outputs and enforcing access policies, so prompts and tool calls fall under one security framework.
Signals and Smart Routing: each request’s intent and complexity is assessed and the request is routed to the best-fit model for your goals. Savings, latency and cache efficiency are tracked in real time, and administrators can automate optimisation simply by specifying target outcomes.
What it means for your organisation
If you are rolling out Claude or similar tools, the questions to ask are: do you know which data goes to the model, can you see which tools your agents reach, and is your AI spend under control? Albisec starts Prisma AIRS projects with an assessment that answers exactly these questions. Get in touch.
Source: Palo Alto Networks Blog, “Agentic Innovation Meets Enterprise Control”, 15 October 2026. Images: Palo Alto Networks.
