Many security leaders head into the cloud armed mostly with tools, practices, skills and ultimately the mental models for how security works that were developed on premise. This leads to cost and efficiency problems that can be solved by mapping their existing mental models to those of the cloud.

When it comes to understanding the differences between on-premises cybersecurity mental models and their cloud cybersecurity counterparts, a helpful place to start is by looking at the kinds of threats each one is attempting to block, detect, or investigate.

Traditional on-premise threats focused on stealing data from databases, file storage, and other corporate resources. The most common defenses of these resources rely on layers of network, endpoint, and sometimes application security controls. The proverbial “crown jewels” of corporate data were not made accessible with an API to the outside world or stored in publicly accessible storage buckets. Other threats aimed to disrupt operations or deploy malware for various purposes, ranging from outright data theft to holding data for ransom.

There are some threats that are specifically aimed at the cloud. Bad actors are always trying to take advantage of the ubiquitous nature of the cloud. One common cloud-centered attack vector that they pursue is constantly scanning IP address space for open storage buckets or internet-exposed compute resources.

As Gartner points out, securing the cloud requires significant changes in strategy from the approach we take to protect on-prem data centers. Processes, tools, and architectures need to be designed using cloud-native approaches to protect critical cloud deployments. And when you are in the early stages of cloud adoption, it’s critical for you to be aware of the division of security responsibilities between your cloud service provider and your organization to make sure you are less vulnerable to attacks targeting cloud resources.

Successful cloud security transformations can help better prepare CISOs for threats today, tomorrow, and beyond, but they require more than just a blueprint and a set of projects. CISOs and cybersecurity team leaders need to envision a new set of mental models for thinking about security, one that will require you to map your current security knowledge to cloud realities.

As a way to set the groundwork for this discussion, the cloud security transformation can start with a meaningful definition of what “cloud native” means. Cloud native is really an architecture that takes full advantage of the distributed, scalable, and flexible nature of the public cloud. (To be fair, the term implies that you need to be born in the cloud to be a native, but we’re not trying to be elitist about it. Perhaps a better term would be “cloud-focused” or doing the security “the cloudy way.”)

However we define it, adopting cloud is a way to maximize your focus on writing code, creating business value, and keeping your customers happy while taking advantage of cloud-native inherent properties—including security. One sure way to import legacy mistakes, some predating cloud by decades, into the future would be to merely lift-and-shift your current security tools and practices into the public cloud environment.

Going cloud-native means abstracting away many layers of infrastructure, whether it’s network servers, security appliances, or operating systems. It’s about using modern tools built for the cloud and built in the cloud. Another way to think about it: You’ll worry less about all these things because you’re going to build code on top of that to help you move more quickly. Abandoning legacy security hardware maintenance requirements is part of the win here. To put another way, security will follow in the steps of IT that has been transformed by the SRE and DevOps revolution.

You can extend this thinking to cloud native security, where some of your familiar tools combine with solutions provided by cloud service providers to take advantage of cloud native architecture to secure what’s built and launched in the cloud. While we talked about the differences between on-prem targeted threats compared to threats targeting cloud infrastructure, here are other vital areas to re-evaluate in terms of a cloud security mental model.

Network security

Some organizations practice network security in the cloud as if it were a rented data center. While many traditional practices that worked reasonably well on-premise for decades, along with many traditional network architectures, are either not applicable in the cloud or not optimal for cloud computing.

However, concepts like a demilitarized zone (DMZ) can be adapted to today’s cloud environments. For example, a more modern approach to DMZ would use microsegmentation and govern access by identity in context. Making sure that the right identity, in the right context, has access to the correct resource gives you strong control. Even if you get it wrong, microsegmentation can limit a breach blast radius.

Becoming cloud native also drives the adoption of new approaches to enterprise network security, such as BeyondProd. It also benefits organizations because it gets them away from traditional network perimeter security to focus on who and what can access your services—rather than where requests for access originated.

Although network security changes driven by cloud adoption can be enormous and transformational, not all areas shift in the same way.

Endpoint security

In the cloud, the concept of a security endpoint changes. Think of it this way: A virtual server is a server. But what about a container? What about microservices and SaaS? With software as a service cloud model, there’s no real endpoint there. All along your cloud security path, users only need to know what happens where.

Here is a helpful mental model translation: An API can be seen as sort of an endpoint. Some of the security thinking developed for endpoints applies to cloud APIs as well. Securing access, permissions, privileged access thinking can be carried over, but the concept of endpoint operating system maintenance does not.

Even with automation of service agents on virtual machines in the cloud, insecure agents may increase risks because they are operating at scale in the cloud. Case in point: This major Microsoft Azure cross-tenant vulnerability highlighted a new type of risk that wasn’t even on the radar of many of its customers.

In light of this, across the spectrum of endpoint security approaches, some disappear (such as patching operating systems for SaaS and PaaS), some survive (such as the need to secure privileged access,) and yet others are transformed.

Detection and response

With a move to the cloud comes changes to the threats you’ll face, and changes to how you detect and respond to them. This means that using on-prem detection technology and approaches as a foundation for future development may not work well. Copying all your on-premises detection tools and their threat detection content won’t reduce risks in the way that most cloud-first organizations will need..

Moving to the cloud provides the opportunity to rethink how you can achieve your security goals of confidentiality, integrity, availability, and reliability with the new opportunities created by cloud process and technology.

Cloud is distributed, often immutable, API-driven, automatically scalable, and centered on the identity layer and often contains ephemeral workloads created for a particular task. All these things combine to affect how you handle threat detection for the cloud environment and necessitate new detection methods and mechanisms.

There are six key domains where threats in the cloud can be best detected: identify, API, managed services, network, compute, and Kubernetes. These provide the coverage needed related to network, identity, compute, and container infrastructure. They also provide specific detection mechanisms for API access logs and network traffic captures.

As with endpoint security, some approaches become less important (such as network IDS on encrypted links), others can grow in importance (such as detecting access anomalies,) while others transform (such as detecting threats from the provider backplane).

Data security

The cloud is changing data security in significant ways, and that includes new ways of looking at data loss prevention, data encryption, data governance, and data access.

Cloud adoption sets you on a path to what we at Google call “autonomic data security.” Autonomic data security means security has been integrated throughout the data lifecycle and is improving over time. At the same time, it makes things easier on users, freeing them from having to define and redefine myriad rules about who can do what, when, and with which data. It lets you keep pace with constantly evolving cyberthreats and business changes, so you can keep your IT assets more secure and make your business decisions faster.

Similar to other categories, some data security approaches wane in importance or disappear (such as manual data classification at cloud scale), some retain their importance from on-prem to cloud unchanged, while others transform (such as pervasive encryption with effective and secure key management).

Identity and access management

The context for identity and access management (IAM) in the cloud is obviously different from your on-premise data center. In the cloud, every person and service has its own identity and you want to be able to control access.

Within the cloud, IAM gives you fine-grained access control and visibility for centrally managing cloud resources. Your administrators authorize who can act on specific resources, giving you full control and visibility to manage cloud resources centrally. What’s more, if you have complex organizational structures, hundreds of workgroups, and a multitude of projects, IAM gives you a unified view into security policy across your entire organization.

With identity and access management tools, you’re able to grant access to cloud resources at fine-grained levels, well beyond project-level access. You can create more granular access control policies to resources based on attributes like device security status, IP address, resource type, and date and time. These policies help ensure that the appropriate security controls are in place when granting access to cloud resources.

The concept of Zero Trust is strongly in play here. It’s the idea that implicit trust in any single component of a complex, interconnected system can create significant security risks. Instead, trust needs to be established via multiple mechanisms and continuously verified. To protect a cloud-native environment, a zero trust security framework requires all users to be authenticated, authorized, and validated for security configuration and posture before being granted or keeping access to cloud-based applications and data.

This means that IAM mental models from on premise security mostly survive, but a lot of underlying technology changes dramatically, and the importance of IAM in security grows significantly as well.

Shared fate for greater trust in cloud security

Clearly, cloud is much more than “someone else’s computer.” That’s why trust is such a critical component of your relationship with your chosen cloud service providers. Many cloud service providers acknowledge shared responsibility, meaning that they supply the underlying infrastructure but leave you responsible for many seemingly inscrutable security tasks.

With Google Cloud, we operate in a shared fate model for risk management in conjunction with our customers. We believe that it’s our responsibility to be active partners as our customers deploy securely on our platform, not delineators of where our responsibility ends. We stand with you from day one, helping you implement best practices for safely migrating to and operating in a trusted cloud.

Get ready to go cloud native

We offer you several great resources to help you prepare for cloud migration, and guide you as you review your current security approaches for signs of on-prem thinking.

Listen to our podcast series where Phil Venables, Vice President, CISO at Google Cloud, and

Nick Godfrey, Director, Financial Services Security & Compliance and member of Office of the CISO at Google Cloud, join me in a discussion on preparing for cloud migration (Podcast 1, Podcast 2). You can deepen your cloud native skills by earning a Professional Cloud Security Engineer certification from Google.

One of the biggest challenges when serving machine learning models is delivering predictions in near real-time. Whether you’re a retailer generating recommendations for users shopping on your site, or a food service company estimating delivery time, being able to serve results with low latency is crucial. That’s why we’re excited to announce Private Endpoints on Vertex AI, a new feature in Vertex Predictions. Through VPC Peering, you can set up a private connection to talk to your endpoint without your data ever traversing the public internet, resulting in increased security and lower latency for online predictions. 

Configuring VPC Network Peering

Before you make use of a Private Endpoint, you’ll first need to create connections between your VPC (Virtual Private Cloud) network and Vertex AI.  A VPC network is a global resource that consists of regional virtual subnetworks, known as subnets, in data centers, all connected by a global network. You can think of a VPC network the same way you’d think of a physical network, except that it’s virtualized within GCP. If you’re new to cloud networking and would like to learn more, check out this introductory video on VPCs.

With VPC Network Peering, you can connect internal IP addresses across two VPC networks, regardless of whether they belong to the same project or the same organization. As a result, all traffic stays within Google’s network.

Deploying Models with Vertex Predictions

Vertex Predictions is a serverless way to serve machine learning models. You can host your model in the cloud and make predictions through a REST API. If your use case requires online predictions, you’ll need to deploy your model to an endpoint. Deploying a model to an endpoint associates physical resources with the model so it can serve predictions with low latency. 

When deploying a model to an endpoint, you can specify details such as the machine type, and parameters for autoscaling. Additionally, you now have the option to create a Private Endpoint. Because your data never traverses the public internet, Private Endpoints offer security benefits in addition to reducing the time your system takes to serve the prediction when it receives the request. The overhead introduced by Private Endpoints is minimal, achieving performance nearly identical to DIY serving on GKE or GCE.  There is also no payload size limit for models deployed on the private endpoint.

Creating a Private Endpoint on Vertex AI is simple.

In the Models section of the Cloud console, select the model resource you want to deploy.

Next, select DEPLOY TO ENDPOINT

In the window on the right hand side of the console, navigate to the Access section and select Private. You’ll need to add the full name of the VPC network for which your deployment should be peered.

Note that many other managed services on GCP support VPC peering, such as Vertex Training, Cloud SQL, and Firestore. Endpoints is the latest to join that list.

What’s Next?

Now you know the basics of VPC Peering and how to use Private Endpoints on Vertex AI. If you want to learn more about configuring VPCs, check out this overview guide. And if you’re interested to learn more about how to use Vertex AI to support your ML workflow, check out this introductory video. Now it’s time for you to deploy your own ML model to a Private Endpoint for super speedy predictions!

Security operations teams are facing a “perfect storm” of challenges from nation-state actors turning their attention to financial crime, to rising uncertainty and potential complexities because of rapidly advancing cloud migration and IoT adoption, to the long-lamented skills shortage. 

Now imagine having to face this trifecta without visibility into your IT and security infrastructure because scale and cost concerns have impeded your ability to ingest and monitor critical telemetry. To help organizations rethink threat detection and broader security team efforts, we showcased at this month’s Google Cloud Security Talks a keynote on why DEI is a cybersecurity imperative, Mandiant’s review of lessons for organizations two years after SolarWinds, how Google Chrome already helps protect your business (and what more you can make do with it,) how best to encourage your organization to pursue Zero Trust, and three key ways in which Chronicle Security Operations can help your organization scale threat detection and actionable outcomes. These include:

1. Operationalizing threat intelligence

Harnessing the power of threat intelligence to acquire context, prioritize risk, and act on detections requires more than just ingesting a set of indicator of compromise (IOC) feeds. Threat intelligence must have breadth and depth of coverage to provide custom details on the tools, tactics, and procedures of threat actors. Explore why part of the “magic” behind Google Cloud’s security is the sheer scale of threat intelligence we can share with our customers. 

2. Driving detection as code

We want to enable organizations that are more advanced in their cybersecurity maturity to build their own custom detections and rules, which means offering our customers a powerful detection-authoring platform. Chronicle Security Operations can help analysts build specific detections around, for example, known bad files or a specific registry key change. We also demonstrated how to create more complex detections, such as flagging process execution patterns or crafting a suspicious-behavior trigger from a specific activity sequence.

3. Leveraging curated detections

With curated detections, which we announced in August, we are putting the power of Google Cloud’s intelligence in the hands of security operations teams everywhere. Our detections offer actionable, ready-to-use threat detection content curated, built, and maintained by Google Cloud Threat Intelligence (GCTI) researchers. These detection sets cover a wide variety of threats for your network and beyond, including attacks such as ransomware, remote-access tools (RAT), infostealers, data exfiltration, and suspicious activity.

You can watch all nine of our Security Talks recorded sessions here, and keep an eye out for our next Security Talks, coming in March 2023.