When an automated service account is compromised, the attacker can move laterally, exfiltrate data, and execute destructive commands across every system the account can reach. The financial and reputational cost of that expanded blast radius can dwarf the original breach.
Current practice leaves agents unchecked
Most Azure teams provision service accounts with long‑lived secrets and embed those secrets in CI pipelines, configuration files, or container images. The agents run with broad permissions that cover multiple subscriptions, resource groups, and databases. Because the connection is made directly from the agent to the target, there is no central point that can observe what command is being sent, mask sensitive fields in responses, or require a human to approve high‑risk actions. Auditing is often limited to a log line that records that a connection was opened, but the actual commands executed remain invisible. This model gives an attacker who steals a single credential the ability to issue arbitrary commands across the entire environment, dramatically inflating the blast radius.
Why identity alone isn’t enough
Introducing non‑human identities, least‑privilege role assignments, and federated OIDC tokens is a necessary step. It ensures that an agent only receives the permissions it needs for a specific job. However, even with those controls in place the request still travels straight to the Azure resource without any mediation. The gateway that could enforce command‑level policies, record each session, or apply inline masking never sees the traffic. Consequently, the system still lacks real‑time approval workflows, fine‑grained audit trails, and the ability to block dangerous commands before they reach the target. The blast radius therefore remains large despite improved identity hygiene.
Putting a gateway in the data path
To shrink the blast radius you need a Layer 7 gateway that sits between the agent and the Azure service. The gateway becomes the only place where traffic can be inspected, authorized, and recorded. hoop.dev provides exactly that data‑path control. It proxies connections to Azure resources, verifies the caller’s OIDC token, and then applies policy checks on every request. Because the gateway holds the credential, the agent never sees the secret. The gateway can enforce just‑in‑time access, require an approval workflow for privileged commands, and record the full session for replay. Inline masking can be applied to responses that contain personally identifiable information, ensuring that even a compromised agent cannot exfiltrate raw data.
How a gateway shrinks blast radius
When hoop.dev is placed in the data path, every command issued by an agent is subject to the following enforcement outcomes: