Two ways agents reach the cloud
The first is from the device. A developer's coding agent runs in a terminal on a laptop, but the laptop holds credentials for cloud accounts, and the agent can call the AWS, Google Cloud or Azure command line tools, kubectl, or a container runtime. One command from a laptop can change production infrastructure. The control point is still the device: the product has to see the command before it runs and decide.
The second is in the cloud. The agent runs in a remote development environment, a cloud-hosted workspace, or as a service built on a cloud AI platform. There may be no company laptop in the path at all. The control point moves to wherever the agent runs.
What vendors document
Bay's capability rules name cloud CLIs, containers and Kubernetes among the actions it can govern, which addresses the first path from the device. Bay's public pages describe it as security for AI agents on enterprise endpoints and do not describe coverage for agents running in cloud workspaces or cloud-hosted apps, which is why it scores 3 out of 10 on our cloud and workspace reach criterion.
Onyx Security lists cloud workloads among the surfaces it covers, alongside browser AI, coding assistants, desktop agents and MCP servers. Zenity covers cloud and homegrown agents and lists integrations with AWS Bedrock, Azure AI Foundry and Google Vertex AI. Noma Security covers endpoint, SaaS and homegrown agents and describes AI-SPM across cloud, SaaS and developer environments. Palo Alto Networks describes agent discovery across cloud, SaaS and endpoints in Prisma AIRS 3.0, and says it is integrating Koi with Prisma AIRS.
Why we weight it at 10%
This site's question covers employee devices and cloud workspaces, so reach beyond the laptop is scored. It carries 10% rather than more because most agent activity we can document today still starts on a device, and because cloud coverage claims on vendor pages are broad and hard to compare. If most of your agents already run in the cloud, raise this weight in the agent risk estimator; the ranking changes.
Questions for a cloud-heavy estate
- Where do your agents run today, and where will they run in a year?
- Which cloud credentials sit on developer devices, and which commands can an agent run with them?
- For agents in cloud workspaces, does the product run inside the workspace, at a gateway in front of the model, or through the cloud platform's own APIs?
- Does the same policy apply on the laptop and in the workspace, or are they separate products with separate rules?
A practical split
Many teams will end up with device control for agents that run on laptops and a separate control for cloud-hosted agents. That is not a failure of either product. Write down which agents each one covers, and check that the actor trail from both lands in the same place.