AWS Security Hub just absorbed two of the biggest blind spots in modern cloud defense: Microsoft Azure resources and production AI workloads. On July 22, the cloud giant unveiled a suite of new capabilities that let security teams monitor Azure virtual machines, container images, function apps, and identities directly within Security Hub—and simultaneously detect threats like cost-harvesting attacks and prompt injection against artificial intelligence services. For Windows administrators and Azure-dependent enterprises, the update is the latest signal that multi-cloud security consolidation is no longer a future aspiration; it’s shipping now.

What Actually Changed

The July update brings three AI-focused features and a native Azure integration to Security Hub, moving the service well beyond its origins as an AWS findings aggregator.

AI Inventory: Keeping Tabs on Every Model and Endpoint

Security Hub now includes an AI Inventory that provides an organization-wide view of AI and machine learning assets. It covers managed services—Amazon Bedrock, SageMaker, and AgentCore—by discovering resources and correlating them with underlying infrastructure (compute, networking, identity, data stores). For self-hosted workloads, the inventory uses signals from Amazon Inspector and Amazon GuardDuty to identify models and inference servers running on EC2 instances or in Elastic Container Registry images. Supported technologies include Ollama, Hugging Face model assets, vLLM, TorchServe, NVIDIA Triton Inference Server, Text Generation Inference, SGLang, llama.cpp, LocalAI, BentoML, Ray Serve, and GPT4All. The system also detects external AI endpoints called by your workloads through GuardDuty DNS telemetry, giving visibility into third-party model APIs your applications might be using without central approval. The AI Inventory is included in the Security Hub Essentials plan at no extra charge, though self-hosted discovery requires Inspector and GuardDuty to be active.

GuardDuty AI Protection: Catching Cost Harvesting and Prompt Injection

GuardDuty expands into AI-specific threat detection with a new AI Protection module. It analyzes CloudTrail events for Bedrock and SageMaker to spot anomalous model invocation patterns—a user suddenly running far more inference requests than usual, for example—and flags potential “cost harvesting” attacks, where stolen credentials are used to consume expensive AI compute at your expense. The service also integrates with Bedrock Guardrails to detect prompt injection attempts, a critical step as AI agents gain access to internal knowledge bases and tools. A 30-day free trial is available for all GuardDuty customers.

GuardDuty Investigation: AI-Assisted Alert Triage

In preview across 10 commercial regions (US East, US West, Canada, and several European and Asia Pacific locations), GuardDuty Investigation uses AI to automatically analyze findings, review 90 days of related account activity, and produce a disposition with confidence level, evidence summary, MITRE ATT&CK technique classification, and recommended actions. It aims to cut the manual labor of separating false positives from genuine threats. AWS emphasizes that outputs require human review—this is analyst assistance, not autonomous response.

Azure Integration: Native Cross-Cloud Monitoring at Last

Security Hub can now discover selected Microsoft Azure resource types: Virtual Machines, Container Registry images, Function Apps, and identities. It evaluates them for misconfigurations, internet exposure, and software vulnerabilities, and checks posture against the CIS Microsoft Azure Foundations Benchmark. Findings appear in the same standardized format as AWS findings, in the same console, alongside the same response workflows. The feature is subject to regional availability: it cannot be created from the Middle East (UAE, Bahrain) or Asia Pacific (Taipei, New Zealand) regions. After a 30-day free trial, Azure resources are priced at the same rates as equivalent AWS resources.

What It Means for You

For Windows and Azure Administrators

If your organization runs Windows Server workloads, SQL Server, or .NET applications in Azure while also using AWS for other services, you can now see Azure misconfigurations and vulnerabilities without leaving the Security Hub console. A vulnerable Azure VM or an exposed function app will show up next to your AWS EC2 and Lambda findings. That promises faster triage and fewer tool-switching interruptions. However, this is not a full replacement for Azure-native security tools like Microsoft Defender for Cloud. Deep Azure policy management, compliance reporting, and platform-specific alerts still require in-console work. Think of Security Hub’s Azure integration as a common risk queue, not a single pane of glass for all Azure security functions.

For Security Operations Teams

Analysts gain a unified finding format across AWS and Azure, which simplifies automation via EventBridge, ticketing, and incident response runbooks. AI-specific alerts close a dangerous gap—cost harvesting can be financially devastating, and prompt injection can expose sensitive data through agent tool use. Investigations will accelerate with AI-assisted triage, but staff must stay disciplined: every AI-generated conclusion needs human validation before containment actions are taken.

For Developers and AI Engineers

The self-hosted AI discovery will likely surface models and inference servers you didn’t know were running. A data science team may be experimenting with Ollama on an EC2 instance without formal approval, or a developer may have integrated a third-party model API inside a line-of-business application. This visibility is the first step toward governing shadow AI. But note that coverage is not universal—it primarily depends on EC2 and ECR scanning, not on Kubernetes, Lambda, or Fargate. You’ll still need additional tooling for full discovery in modern, cloud-native stacks.

For IT and Business Leaders

The expansion strengthens the case for consolidating cloud security operations around AWS Security Hub, especially if your organization already uses it heavily. However, the new features draw on multiple services—GuardDuty, Inspector, and potentially partner integrations—which all contribute to your monthly bill. Model the total cost of ownership before committing, and validate that the resource types and regions covered align with your actual Azure footprint. The Security Hub Extended program, which brings curated partner solutions into the same interface, adds another layer of potential consolidation and cost.

How We Got Here

AWS Security Hub launched in 2019 as a centralized place to view findings from AWS services like GuardDuty, Inspector, and Macie. For years it was an AWS-only affair. Then, earlier in 2026, AWS introduced Security Hub Extended, a framework for integrating partner security products across nine categories—endpoint, identity, email, network, data, browser, cloud, AI, and security operations—using the Open Cybersecurity Schema Framework. In June, AWS quietly previewed Azure resource monitoring. The July announcement brings all these threads together, bolting on AI-specific protections just as generative AI workloads move from experimentation to production at scale. The move responds to twin realities: most large enterprises run in multiple clouds, and AI assets—models, agents, endpoints—have become too valuable and too risky to leave unmonitored.

What to Do Now

  1. Map your AI estate – Identify all Bedrock, SageMaker, and AgentCore deployments across AWS accounts. List any self-hosted inference workloads, even if they’re experimental, and catalogue external AI APIs your applications call.
  2. Enable AI Inventory – Turn on Security Hub’s AI Inventory (no extra cost in Essentials). Confirm that managed services appear correctly. Ensure Inspector and GuardDuty are active to populate self-hosted and external endpoint views.
  3. Pilot GuardDuty AI Protection – Start with accounts that actively use Bedrock or SageMaker. Establish normal invocation baselines so you can distinguish real anomalies from routine spikes. Create playbooks for cost-harvesting and prompt injection alerts.
  4. Integrate Azure selectively – Begin with subscriptions hosting business-critical VMs, container registries, function apps, and identities. Compare Security Hub findings with what your Azure-native tooling already shows to identify coverage gaps.
  5. Define a cross-cloud triage process – Agree on severity definitions, ticket routing, and escalation paths that work for both AWS and Azure findings. Train analysts to treat AI-generated investigation conclusions as starting points, not final verdicts.
  6. Measure coverage, not just adoption – Track which accounts, regions, resource types, and AI services are actively monitored. Document gaps—for example, Azure Kubernetes Service clusters that aren’t discovered—and plan compensating controls.

Outlook: What to Watch Next

The GuardDuty Investigation preview will likely expand to more regions before general availability. Expect AWS to gradually increase the set of Azure resources covered—managed databases, Kubernetes, and networking services are obvious next targets. On the AI front, detection capabilities will mature as attackers refine techniques specific to large language models and agentic workflows. Meanwhile, Microsoft will almost certainly respond by deepening its own multi-cloud security story in Defender for Cloud. For now, the message for Windows shops is clear: cloud and AI security are converging, and the tools to manage them are becoming less provider-bound.