Explore our new Cybersecurity Terms Glossary, explained by our experts.
Cybersecurity Terms Glossary
A service mesh gives every workload a cryptographic identity, encrypts traffic between services automatically, and enforces who may call whom — provided its own control plane is protected.

What Is Service Mesh Security?
A service mesh is an infrastructure layer that intercepts the traffic between the services in a distributed application and applies encryption, identity, authorisation, and telemetry to it without the applications needing to implement any of that themselves. Istio, Linkerd, Consul, and Cilium’s mesh all work on the same principle: a data plane of proxies alongside each workload, and a control plane that configures them. Service mesh security is the question of what the mesh gives you, what it does not, and what it adds to your attack surface.
The useful shift a mesh makes is from network location to workload identity. Without one, a service decides whether to trust a caller based on its IP address or the fact that it reached the port at all. With one, each workload receives a short-lived X.509 certificate encoding an identity — typically a SPIFFE identifier such as spiffe://cluster.local/ns/payments/sa/ledger — issued by the mesh certificate authority and rotated automatically, often hourly.
Mutual TLS then authenticates both ends of every connection, and authorisation policy is written against identities rather than subnets. A rule saying that only the ledger service account in the payments namespace may call the settlement API on POST /v1/transfer survives pod rescheduling, IP reassignment, and cluster scaling, which no firewall rule does. This is the practical mechanism behind most zero trust architectures inside a cluster, and it maps directly onto the identity-based access control described in NIST SP 800-207.
Meshes ship permissive by default so that adoption does not break anything, and estates frequently stay there.
PeerAuthentication defaults to PERMISSIVE, which accepts both plaintext and mutual TLS. Migration needs it; production does not. Moving to STRICT per namespace, and then mesh-wide, is the change that makes the encryption claim true.AuthorizationPolicy in a namespace, everything is permitted. Apply a deny-all policy first and add explicit allows, exactly as you would with network segmentation.One confusion is worth naming plainly: mutual TLS proves who the caller is; it does not decide what they may do. A mesh in strict mode with no authorisation policies encrypts everything and authorises everyone.
Adding a mesh adds a control plane that can reconfigure routing and policy for every service, plus a certificate authority that can mint any identity in the trust domain. Compromise of that control plane is total.
The track record is instructive. CVE-2021-31920 allowed an Istio authorisation policy to be bypassed through path confusion with multiple slashes. CVE-2022-21654 let a workload with no authorisation configure gateways it should not have controlled. CVE-2023-27496 and related issues caused the Istio control plane to crash on malformed input, taking configuration distribution with it. None of these were exotic; all required prompt upgrades.
So treat the mesh as privileged infrastructure. Restrict who may create or modify mesh custom resources through role-based access control, because an engineer with permission to write an AuthorizationPolicy can grant themselves access anywhere in the mesh. Protect the certificate authority key — integrating with an external authority such as HashiCorp Vault or a cloud key management service removes it from the cluster entirely. Keep the control plane in its own namespace with tight network policy, and patch promptly; mesh components move quickly and support windows are short.
For a handful of services, probably not. The honest comparison is against the alternatives: application-level TLS with a shared certificate authority, Kubernetes NetworkPolicy for coarse segmentation, and an API gateway at the edge. Those cover a great deal for less operational cost.
A mesh earns its place when you have enough services that per-service TLS becomes unmanageable, when you need identity-based authorisation rather than network-based, when regulators ask for demonstrable encryption in transit between internal components, or when you want consistent telemetry across a polyglot estate. Linkerd is markedly simpler than Istio for teams whose requirement is mutual TLS and basic policy; Istio’s ambient mode removes per-pod sidecars and reduces overhead considerably, at the cost of a newer and less-proven architecture.
Whichever you choose, verify the result rather than assuming it. Our cloud security services include testing whether mesh policy actually holds when a workload inside the cluster starts behaving badly, which is the only way to find the namespace somebody excluded from injection last spring.
No, and they work best together. NetworkPolicy operates at layers three and four and is enforced by the CNI plugin, so it still applies to pods with no sidecar and to traffic that never reaches the proxy. Mesh authorisation operates at layer seven with workload identity. Use NetworkPolicy as the coarse floor that nothing escapes, and mesh policy for fine-grained, identity-aware rules above it.
It supplies authentication and encryption, which are prerequisites, but zero trust also requires authorisation decisions per request, least privilege, and continuous verification. A mesh in strict mTLS mode with no authorisation policies has simply encrypted an open network. Pair strict mode with default-deny authorisation policies, scoped service accounts, and monitoring of denied requests before claiming the architecture is zero trust.
Sidecar proxies typically add a few milliseconds per hop and consume meaningful CPU and memory across a large cluster, since every pod gains a proxy. Mutual TLS handshakes add further overhead, mitigated by connection reuse. Sidecar-less approaches such as Istio ambient mode and eBPF-based meshes reduce this substantially. Measure against your own latency budget rather than relying on published benchmarks.
Very few people. Permission to create or edit mesh custom resources is effectively permission to redirect traffic and grant access anywhere in the trust domain, so treat it as a privileged role rather than an ordinary namespace permission. Review policy changes through code review in a GitOps workflow, alert on direct modifications to the cluster, and keep control plane namespaces separate from application namespaces.
You May Be Interested In These
We use a few cookies that are needed to make this website work, and they are always on. We would also like to set analytics cookies, to see which pages are useful; functional cookies, so that our live chat works; and advertising cookies, to measure our campaigns. We will not set any of those until you say yes, and you can change your mind at any time from the Cookie Settings link in the footer.
Choose the cookies you are happy for us to use. Strictly necessary cookies are always on; everything else is off until you turn it on.
These cookies keep the website working: they hold your security checks, your form submissions and the choice you make here. The site cannot run without them, so they are always on.
These let our live chat open and remember your conversation, and let the newsletter sign-up work. Switch them off and the chat window will not appear.
These tell us which pages are read and how people find us, so that we can improve the site. The reports are about groups of visitors, never about you personally.
These let us measure our advertising and show you our adverts on other websites, such as LinkedIn, Google and Meta. Switch them off and you will still see adverts, but they will be less relevant.