A pod started failing, and our secrets were no longer loading. That’s when we found the Sidecar was injecting a broken GPG agent, and everything downstream was on fire.
GPG sidecar injection is a quiet risk until it takes down deployments. It happens when an automated sidecar container mounts secret material—like GPG keys—into a workload at runtime. Most setups rely on sidecars for secure secret delivery. But if the injection is misconfigured, stale, or insecure, it can break trust chains, cause signature verification failures, or leave keys exposed in a running pod.
The lifecycle of injected keys is critical. GPG agents in sidecars often load on container start, but don’t refresh properly. Long-lived pods can keep expired keys in memory, or worse, retain revoked ones. Without explicit reloading, builds and deploys will fail silently until something forces a restart.
Attack surfaces are easy to miss. Unisolated temp volumes hold decrypted key material. Incomplete cleanup scripts let secrets linger after a pod termination. Mount points become predictable when standardized across environments, giving insiders a simple path to locate and copy private key data. Injection also risks cross-namespace bleed if volumes are mis-scoped.