Navigating the AI Sprawl

Nov 19, 2026
5 minutes

Rapid adoption of SaaS agents, AI coding assistants, and enterprise agents is transforming operations and software development. However, employees and developers rapidly introducing new models, skills, plugins, and agents often bypass traditional security reviews, causing unprecedented sprawl. As AI agents evolve into connected, autonomous systems spanning endpoints, browsers, SaaS applications, and the cloud, they introduce increasingly complex attack surfaces. 

To safely adopt AI at scale, organizations need a cohesive strategy that brings this massive sprawl under control without slowing down innovation. To help build on that strategy with Prisma AIRS, today, we are announcing solutions that deliver:

  1. Precise Identity Controls via Agent Identity Security
  2. Frictionless Developer Security via Direct Integrations
  3. Automated Threat Defense via New Enhancements

Redefining Control in Agents 

Controlling AI sprawl starts with identifying every actor in your environment. IDC projects a 40-fold increase in enterprise AI agents by 2029, creating a massive new class of privileged identity. Currently, most agents rely on human identities or service accounts, automatically inheriting excessive permissions. Because non-deterministic agents cannot be trusted the same way as an organization's human employees, this rapid expansion requires a new approach to identity and access management.

To establish a foundation for secure scale, Prisma AIRS now provides comprehensive Agent Identity Security specifically designed for the enterprise agent ecosystem in order to:

  • Reduce Shadow AI: Maintain ownership across the agent lifecycle with a centralized repository for continuous visibility into active agents and permissions.
  • Prevent Impersonation: Assign unique, verified identities to prevent credential theft and avoid inheriting overly broad human privileges.
  • Contain Blast Radii: Enforce runtime least-privilege access, ensuring agents only reach required resources and preventing lateral movement if compromised.
  • Accelerate Investigations: Log agent and user identities side-by-side to deliver end-to-end traceability and simplify compliance reporting.

By issuing agents their own verified identities, every tool execution is explicitly scoped to what the agent needs in that exact moment.

Securing AI-assisted Coding with Direct Integrations

As agents expand via models, skills, MCP servers, and extensions, vulnerabilities or malicious components can arise. To help safeguard agents and govern traffic end-to-end, we are introducing three key direct integrations.

Native OpenAI Codex Integration

Enable developers to safely scale AI-assisted development without risking data exposure or operational friction. The native integration of Prisma AIRS Runtime API with OpenAI Codex delivers enterprise-grade guardrails directly into automated coding workflows. Operating synchronously at the platform level, it protects proprietary source code, credentials, and sensitive data while intercepting runtime threats before model inference occurs allowing developers to maintain maximum velocity and security teams to maintain full visibility and automated control. This integration:

  • Enables seamless admin-level security enforcement without requiring client hooks or altering developer tools.
  • Intercepts sensitive secrets, credentials, PII, and proprietary code in real time before prompt submission.
  • Proactively inspects developer inputs to block prompt manipulation attacks, malicious code patterns, and unsafe URLs.

Expanded Integration with Cortex Agentic Endpoint Security

Secure developer environments without slowing down innovation. By combining endpoint intelligence from Cortex AES with Prisma AIRS, security teams gain actionable context over local coding agents like Cursor, Claude Code, and Codex, enabling friction-free adoption alongside robust threat protection. This integration empowers teams to:

  • Gain complete, automated oversight of hidden AI artifacts across all endpoints.
  • Detect risks preemptively by intercepting prompts, MCP tool calls, and web connections prior to off-device transmission.
  • Neutralize high-risk agent actions inline to maintain uninterrupted development operations and tool uptime.

AI Gateway Integration with Prisma SASE

Streamline deployment and eliminate infrastructure overhead by delivering high-performance, automatically scalable AI security right where your teams work. With Prisma AIRS AI Gateway integrated into Prisma SASE, developers seamlessly and securely connect to public or private MCP servers and LLMs using existing network paths. Organizations gain complete visibility and control over model interactions, MCP calls and agent-to-agent traffic. This gives organizations the ability to:

  • See your AI usage and optimize the cost through AI Gateway in Strata Cloud Manager.
  • Secure all AI interactions through Prisma AIRS Runtime Security.
  • Scale your AI agents deployment across the organization with a fully managed solution

Additional Prisma AIRS Platform Enhancements

We are also constantly improving our platform to stay ahead of threats, which is why we have recently launched four powerful capabilities in Prisma AIRS enabling you to:

  • Automate Defenses via Informed Runtime Policies: Automatically convert risk insights into active runtime defenses by combining deep threat analysis with clear exposure visibility across your AI models.
  • Scale Safely in Regulated Markets with FedRAMP Moderate Support: Enable government and highly regulated organizations to confidently deploy AI with compliant, built-in protection against prompt injection and data loss.
  • Maintain Business Continuity via Post Processing Workflow: Keep legitimate workflows running and eliminate false positives through secondary AI-driven threat validation before any block is applied.
  • Preempt Attacks with Multi-Turn AI Red Teaming: Proactively discover hidden vulnerabilities and harden applications before production by simulating complex, multi-turn attack strategies like Crescendo and GOAT.

To explore these capabilities in greater depth and see live demonstrations of them in action, join us for our AI Security Innovation Showcase webinar. The session features an insightful virtual chat with OpenAI, discussing how we can collectively advance enterprise security, protect autonomous workflows, and safely unleash the next generation of AI innovation.