Dev48
Language
  • About
  • Services
  • Industries
  • Technologies
  • Articles
  • Contacts
Book a call
    Home/Articles/The ai attack surface doesnt end at the model
Dev48
info@dev48.org+375 29 320-40-96WhatsApp@isky_by

© 2026 · All rights reserved.

The AI Attack Surface Doesn't End at the Model

Source: Palo Alto Networks Blog•November 22, 2026 at 12:03 AM

The AI Attack Surface Doesn't End at the Model

Discover how Cortex Cloud 2.3 expands AI security across AWS and Google Cloud with visibility into AI agents, infrastructure, identities, and security risks.

Building an AI agent takes more than a model.

An agent might retrieve sensitive information from a knowledge base, authenticate through a service identity, or connect to external tools to complete a task. Each interaction depends on resources with their own permissions, configurations, and potential security risks.

Cloud providers are expanding their AI offerings beyond model development, introducing managed agent platforms and infrastructure that make it easier to build, deploy, and scale AI applications. But as these environments grow, so do the potential points of exposure.

For security teams, knowing which AI models are running is no longer enough. They also need to understand what agents can access, how supporting infrastructure is configured, and where security risks may emerge across the AI ecosystem.

But as developers spin up new agents, workspaces, and services, how much of that infrastructure can security teams actually see? An overlooked configuration or overprivileged identity could expose sensitive data long before anyone realizes there's a problem.

The Expanding AI Ecosystem Introduces New Security Challenges

A cloud inventory might show a compute instance, storage bucket, or service identity. But can it tell you which AI agents are running, what data they can access, or how their supporting infrastructure is configured?

Traditional cloud security tools can identify the underlying resources, but often lack the context to recognize agent runtimes, model endpoints, and development workspaces as distinct AI assets.

An agent operating with excessive permissions could access resources beyond its intended scope. A misconfigured notebook could expose sensitive data. Even when the underlying infrastructure is monitored, these risks can go unnoticed without visibility into the AI applications and services running on it.

Securing the AI ecosystem starts with bringing its hidden infrastructure and potential exposures into view.

What’s New: Unified AI Security Across AWS and Google Cloud

Cortex Cloud is closing the visibility gap in the managed AI ecosystem. With the release of Cortex Cloud 2.3, we are extending AI Security Posture Management (AI-SPM) to the core infrastructure layers where enterprise agents live and breathe.

Security teams can now achieve a complete AI Bill of Materials (AI-BOM) across their multi-cloud environments, mapping the entire lineage of an agent - from the underlying foundation models and datasets to the active runtimes and connected tools.

Key Capabilities in this Release:

  • Expanded Visibility for AWS Bedrock AgentCore: Automatically discover and secure the managed infrastructure supporting your AWS agents, including Agent Runtimes, Gateways and Agent Tools.
  • Deep Security for GCP Gemini Enterprise Agent Platform: Gain granular insights into Agent Identities and Credentials, Gateways and MCP Servers to ensure fine-grained access control and data governance.
  • Inactive Agent & Endpoint Detection: Identify stale or experimental agents and model endpoints that are no longer in use, allowing you to proactively reduce your attack surface and eliminate "AI Sprawl."
  • Enhanced SageMaker Support: Extended discovery for SageMaker Domains, Spaces, and Notebooks, securing the AI development lifecycle long before models reach production.

Build a More Complete Picture of AI Security Risk

As organizations expand their use of AI across AWS and Google Cloud, securing these environments requires visibility beyond models and agents alone.

With expanded AI-SPM coverage in Cortex Cloud 2.3, security teams can discover the infrastructure, identities, and connected resources supporting their AI applications, identify risky configurations, and better understand potential exposures across their environments.

By bringing these assets together in a unified AI Bill of Materials (AI-BOM), Cortex Cloud helps organizations identify and address security risks as their AI environments continue to evolve.

See Cortex Cloud AI Security in Action

Want to see how Cortex Cloud uncovers security risks across your AI ecosystem?Request a demo to explore AI asset discovery, risk detection, and security posture management across AWS and Google Cloud.

← All articles

What this article says

More in Cybersecurity

All →
Signal restricts mobile screenshots as it confronts ‘existential threat’ of AI surveillanceПресса
Signal

Signal restricts mobile screenshots as it confronts ‘existential threat’ of AI surveillance

Navigating the AI Sprawl
CyberArk

Navigating the AI Sprawl

Real-Time Policy Enforcement: Palo Alto Networks Firewalls Integrate with AWS Transit Gateway Context Sharing
CyberArk

Real-Time Policy Enforcement: Palo Alto Networks Firewalls Integrate with AWS Transit Gateway Context Sharing

Zscaler stock jumps 18% on report that CrowdStrike in talks to buy security vendorПресса
Zscaler

Zscaler stock jumps 18% on report that CrowdStrike in talks to buy security vendor

Founders in Focus: Hagai Shapira of Daylight
Okta

Founders in Focus: Hagai Shapira of Daylight

Why the Harness Matters: Early Lessons From the Evo Code Scanning Agent Preview
Snyk

Why the Harness Matters: Early Lessons From the Evo Code Scanning Agent Preview

More from CyberArk

Navigating the AI Sprawl
CyberArk

Navigating the AI Sprawl

Real-Time Policy Enforcement: Palo Alto Networks Firewalls Integrate with AWS Transit Gateway Context Sharing
CyberArk

Real-Time Policy Enforcement: Palo Alto Networks Firewalls Integrate with AWS Transit Gateway Context Sharing

Secure SI by Design Consortium Officially Approved by U.S. Government
CyberArk

Secure SI by Design Consortium Officially Approved by U.S. Government

"That Wasn't Me, Officer!"
CyberArk

"That Wasn't Me, Officer!"

Palo Alto Networks Brings Continuous AI Defense and Prisma AIRS to the OpenAI Marketplace
CyberArk

Palo Alto Networks Brings Continuous AI Defense and Prisma AIRS to the OpenAI Marketplace

Reinventing Endpoint Security with Cortex XDR One
CyberArk

Reinventing Endpoint Security with Cortex XDR One