AWS turns Security Hub into an AI and multicloud security control plane

By extending Security Hub to Microsoft Azure and adding artificial intelligence-specific protections, Amazon Web Services Inc. is positioning its security stack as the foundation for securing enterprise AI at scale.

Amazon Web Services’ latest Security Hub updates announced earlier this month acknowledge two realities its customers already live with every day: AI is now the fastest-growing attack surface in the enterprise, and almost every serious AI initiative is, or soon will be, multicloud. The result is a Security Hub release that extends to Microsoft Azure and delves into AI workloads on AWS, with the clear goal of turning the service into a unified control plane for cloud and AI security.

Security Hub, which began as a cloud security posture management or CSPM tool for AWS, has steadily evolved into the aggregation point for detections across native services and partners. The new release expands that role. I recently discussed the update with Michael Fuller, director of security services at AWS, who explained, “Customers have gone multicloud, and have been pretty loud and clear: Don’t make me buy a different tool when I already have Security Hub.”

What AWS announced

The Security Hub announcement bundles four main capabilities:

• Azure resource monitoring
• GuardDuty AI Protection for AI workloads
• GuardDuty AI-powered investigations (preview)
• Security Hub AI inventory

Regarding Azure, Security Hub can now discover and monitor Azure Virtual Machines, container images in Azure Container Registry, Function Apps and identities. It evaluates those resources for misconfigurations, internet exposure and software vulnerabilities, including checks aligned with the CIS Microsoft Azure Foundations Benchmark. The findings appear side by side with AWS findings, using the same format, prioritization logic and automation workflows. That means a security engineer can finally see AWS and Azure posture through a single lens instead of juggling separate consoles and rule sets.

Under the covers, AWS is leveraging one of its core security primitives: AWS Config. It has quietly gone multicloud and now powers near-real-time, change-triggered evaluation of Azure resources, rather than the 24-hour polling cycle typical of many third-party CSPM tools. Fuller emphasized that customers don’t have to wire any of this together themselves. “We’ve replicated the one-click experience customers love from AWS,” he said. “You make a tenant-level decision in Azure, give us read-only access once, and we do the rest under the covers.”

Pricing has been deliberately simplified. Security Hub consolidates what used to be multiple line items into a per-resource-per-month model and charges only for four resource types — virtual machiness, container images, functions and identities — across both AWS and Azure. Fuller framed the philosophy as making Security Hub a “no-brainer” alternative to building your own detection and posture pipeline from cloud primitives, rather than a race to the absolute lowest line-item price.

AI workload protection: The strategic move

In my opinion, the most significant part of this release is squarely in the AI stack. AWS is introducing GuardDuty AI Protection, AI-powered investigations and a new AI inventory within Security Hub to help enterprises secure model-driven workloads on services such as Amazon Bedrock and SageMaker.

GuardDuty AI Protection targets threats unique to AI services, including anomalous model invocations that signal abuse of an inference endpoint, prompt-injection attempts (via integration with Bedrock Guardrails),and what AWS calls “cost harvesting” attacks. Cost harvesting is essentially the AI-era version of crypto-mining abuse: an attacker gains access to credentials, spins up large volumes of inference and converts your AI budget into free compute. Fuller noted that this is not an entirely new category so much as a new expression of an old pattern. From the earliest days of the cloud, compromised accounts were abused to run unauthorized workloads; AI inference is simply the latest, and sometimes most lucrative, target.

GuardDuty AI-powered investigations, currently in preview, directly address the number one pain point security teams cite: alert fatigue. When GuardDuty raises a finding, the investigation engine automatically retrieves related activity, context on the affected resources and relevant threat intelligence. It then generates a confidence score, MITRE ATT&CK mapping, supporting evidence and recommended actions — such as whether to suppress, contain or fully investigate the issue. Analysts still make the final call, but they no longer must manually stitch together logs and events for every alert.

Critically, AWS isn’t throwing a generic large language model at the problem. Fuller noted that GuardDuty has been live since 2017, and AWS has nearly a decade of experience helping customers and its own internal teams triage findings. That history has been used to teach the AI investigator which questions to ask and what context to look for, whether that’s a Palo Alto Networks Inc. firewall behind an EC2 instance or traffic routed through a customer’s on-premises data center.

The new AI inventory in Security Hub ties everything together by providing a single view of AI assets — Bedrock models, SageMaker endpoints and agents built with Amazon AgentCore — and mapping them to the underlying compute, networking, IAM and data stores they depend on. That makes it much easier for security teams to answer basic but critical questions: What AI workloads do we actually have? Where are they running? If one is compromised, which systems and data are in the blast radius?

How this helps accelerate enterprise AI

Most organizations aren’t slowing AI pilots because they lack models or GPUs; they’re slowing down because risk and compliance teams don’t have guardrails they trust. This release is AWS’ attempt to move “secure by default” for AI workloads from slide decks to something closer to operational reality.

There are a few concrete accelerators here.

• A unified view across clouds. If your security operations can see misconfigurations and vulnerabilities across AWS and Azure in a single console, using a single set of workflows, it becomes much easier to sign off on hybrid AI architectures spanning multiple clouds. That directly shortens the approval loop for new AI services.
• AI-aware threat detection. Treating AI services as first-class citizens in GuardDuty prevents AI from becoming a blind spot. Anomalous model activity, cost harvesting and prompt injection are surfaced alongside more traditional network and endpoint threats, rather than in a separate, under-instrumented silo.
• Triage at AI speed. AI-powered investigations are less about replacing analysts and more about compressing the time it takes to distinguish real attacks from background noise. Fuller acknowledged that customers don’t want fewer alerts; they want all of them, but ranked and explained. The investigations preview is an early step in that direction.
• An integrated partner ecosystem. Security Hub Extended now brings 21 curated partner solutions across nine categories into a more unified experience, using OCSF to normalize findings. That gives enterprises a way to keep best-of-breed tools like CrowdStrike, Okta, Cyera and Splunk in the mix without owning all the integration work. As Fuller put it, customers are tired of the “false choice” between a single-vendor suite that isn’t best-of-breed and a best-of-breed stack they have to wire up themselves.

For enterprises trying to industrialize AI, that integration layer is increasingly where projects stall. When cloud, identity, endpoint and AI signals all land in a common findings plane, it becomes much easier to make AI a first-class citizen of existing security operations rather than an exception process.

What IT and security professionals should do next

For IT professionals and security architects, the question isn’t whether this is interesting — it’s what to do with it over the next 12 to 18 months.

First, if you’re already invested in GuardDuty and Security Hub, treat AI protections and the AI inventory as table stakes for any Bedrock or SageMaker deployment. Make enabling GuardDuty AI Protection and registering services in the AI inventory policy requirements before AI workloads move into production. That gives risk and audit teams a consistent framework for assessing AI risk with tools they already know.

Second, if you’re an AWS-first shop with a growing Azure footprint, you should pilot Security Hub’s Azure monitoring on a representative subscription and compare it with your existing CSPM tool, especially regarding latency and operational friction. You don’t need to rip and replace overnight, but you should have a clear target state for where cloud posture management lives. Increasingly, it will be difficult to justify a fractured approach that looks different in each cloud.

Third, use the AI-powered investigations preview to address alert fatigue in a measured way. Start by enrolling a subset of GuardDuty findings, track mean time to triage before and after, and use the confidence scores and recommendations to refine your suppression rules and playbooks — not to bypass analyst judgment. This is a low-risk way to get your SOC comfortable with AI “in the loop” while keeping humans firmly on the hook for final decisions.

Finally, zoom out to assess your longer-term SIEM and data strategy. AWS’s direction of travel — consolidated findings in Security Hub, shared data stores for security and observability, and partner tools querying common data — chips away at the traditional model in which every security tool insists on owning all of your raw data. You don’t need to blow up your SIEM today, but it’s worth aligning your next renewal with a plan for how Security Hub, GuardDuty and your partner tools will share data, rather than each demanding its own silo.

Final thoughts

AWS is doing what customers have been asking for: using its weight to collapse security seams across clouds and between traditional workloads and AI. For organizations trying to move faster on AI without losing control of risk, Security Hub’s new multicloud and AI-focused capabilities are worth more than a casual look — this is a blueprint for how cloud security will need to work in the AI era.

Zeus Kerravala is a principal analyst at ZK Research, a division of Kerravala Consulting. He wrote this article for SiliconANGLE.

Image: AWS