AWS is drawing attention to a security issue that matters well beyond enterprise platform teams: some common protocols were not designed with modern multi-tenant cloud environments in mind. In its recent security guidance for Amazon EKS, AWS said protocols such as Android Debug Bridge (ADB) and WebRTC can expose organizations to risk because they may use unencrypted channels by default or lack built-in enterprise authentication. In a shared Kubernetes cluster, that becomes more than a technical detail. If cross-namespace traffic is not restricted, one tenant’s pod may be able to reach another tenant’s open ports by default. For businesses building internal platforms, remote device labs, customer demos, AI testing environments, or any shared service on Kubernetes, the message is practical: isolation at the namespace level is not enough on its own. Network controls, identity, certificate management, and relay choices all affect whether your environment is merely organized or actually protected. For PostPilot Daily readers, this story also reflects a broader operational lesson. As teams publish more apps, services, and AI-assisted workflows faster than ever, audience trust increasingly depends on invisible infrastructure decisions. Security architecture may not make social posts or landing pages better directly, but it shapes reliability, compliance posture, and the confidence you can bring to customers, clients, and partners.
What to know
- AWS says some protocols, including ADB and WebRTC, can create security gaps in multi-tenant EKS clusters when used with default behavior.
- In multi-tenant Kubernetes, pods may reach other tenants’ open ports by default unless network policies restrict cross-namespace traffic.
- AWS described a reusable architecture based on namespace-scoped mTLS gateways, managed TURN relay through Kinesis Video Streams, and JWT identity chains tied to an enterprise identity provider.
- AWS explicitly warned against an older pattern that routed traffic through a bastion host and the Kubernetes API server for low-latency environments.
- The practical lesson for platform teams is to treat protocol security, identity, and network isolation as part of product quality, not just infrastructure hygiene.
Why this AWS EKS warning matters to growing teams
The AWS guidance focuses on multi-tenant Kubernetes, where different users, teams, or workloads share a cluster while remaining logically separated. That setup is attractive because it can reduce operational overhead and make scaling easier. But it also increases the consequences of weak defaults. If a protocol assumes trust or leaves encryption and authentication up to implementers, those gaps can become tenant-to-tenant exposure risks.
AWS highlighted that ADB uses an unencrypted channel by default and requires a proxy or custom API for authentication. It also noted that similar TCP services, including VNC, can have credentials that are local to the protocol rather than tied to a central identity system. WebRTC has a different profile: media is encrypted, but signaling still requires teams to handle authentication and encryption choices themselves.
Analysis: For smaller organizations, this matters because technical debt often enters through convenience. A team may launch a shared development environment, QA device farm, or demo cluster before fully standardizing identity and network policy. That can work for a while, until more tenants, more contractors, or more customer-facing use cases turn a temporary shortcut into a structural risk.
- Shared clusters save effort, but shared infrastructure raises the cost of weak protocol defaults.
- Namespace isolation alone does not prevent all unwanted lateral access.
- Authentication, encryption, and routing decisions affect both security and operational complexity.
What AWS says is the core security gap
The core problem described by AWS is straightforward: in a multi-tenant EKS cluster, pods can reach other tenants’ open ports by default unless teams configure network policies to stop that traffic. That means an insecure or weakly authenticated protocol does not just expose a single workload. It can expose interactions across tenants inside the same cluster.
AWS also emphasized that fixing each protocol one by one does not scale. If every team has to bolt on its own authentication layer, relay design, and access pattern, the organization ends up with custom exceptions instead of a reusable security model. The architecture in the post was presented as a way to standardize controls around multiple protocols using common components.
Recommendation: If your business runs a shared Kubernetes environment, review not just internet-facing ingress but also east-west traffic between namespaces. Many teams focus heavily on external threats while underestimating what becomes reachable internally by default.
- Pods may be able to access open ports across tenants unless blocked by policy.
- Protocol-by-protocol fixes create maintenance burden.
- Reusable security layers are easier to govern than many custom workarounds.
Put the idea into practice with PostPilot Daily’s content planner and publishing workspace.
The architecture AWS recommends instead
AWS described an approach built around three reusable components: namespace-scoped mutual TLS gateways, managed TURN relay through Amazon Kinesis Video Streams, and JWT identity chains connected to an enterprise identity provider. The post said the design assumes namespace isolation is already configured, along with an enterprise identity provider, AWS Private CA, cert-manager, the AWS Private CA issuer plugin, and permission to deploy Envoy Gateway.
For ADB traffic, AWS said each namespace runs its own Envoy Gateway instance. Short-lived mTLS certificates are issued per namespace using AWS Private CA through cert-manager. The custom API validates the user’s JWT against the enterprise identity provider, confirms the namespace assignment, and returns connection details and the namespace certificate over HTTPS. The direct connection then uses mTLS through Envoy to the target pod.
AWS said each certificate has a 12-hour lifetime, with renewal handled automatically by cert-manager. Envoy picks up renewed certificates without a pod restart. The company noted that the ADB path does not have AWS WAF protection and instead relies on mTLS, network policies, and namespace isolation. It also pointed out that the hop between Envoy and the pod is not protected by mTLS in this design, suggesting sidecar proxies or pod-level certificates for environments with stricter compliance requirements.
- Namespace-scoped Envoy Gateway is used for mTLS-authenticated TCP proxying.
- AWS Private CA and cert-manager automate certificate issuance and rotation.
- JWT authentication proves user identity, while mTLS ties the connection to the correct namespace.
- AWS says default-deny NetworkPolicy should be in place before deploying the gateway and certificates.
Why AWS moved away from the earlier bastion-based design
AWS included a previous architecture for comparison and clearly said not to implement that design. In the older model, traffic flowed through a bastion host to the EKS control plane, with a custom API handling authentication and authorization. Users authenticated through a custom CLI, and access relied on a two-hop port-forwarding setup from the user’s machine to the bastion and then to the target pod.
The issue was not only elegance. AWS said routing all traffic through the Kubernetes API server created a throughput bottleneck during concurrent sessions, even though the cluster remained private. For low-latency use cases, that pattern became a practical limitation.
Analysis: This is a good reminder that secure-looking architecture is not always production-ready architecture. If every access path depends on a chokepoint, scaling user activity can degrade performance or reliability. That matters for any team supporting internal creators, QA testers, sales engineers, or distributed operators who depend on smooth access to shared systems.
- AWS marked the older bastion pattern as deprecated.
- Port-forwarding through the Kubernetes API server became a bottleneck under concurrent load.
- Performance constraints can turn access-control designs into operational liabilities.
Why the managed TURN choice matters for operations and cost
For WebRTC, AWS said a TURN relay is needed when direct peer-to-peer connections cannot be established behind a NAT Gateway. The earlier deployment used STUNner, but AWS reported three issues: lack of native integration with AWS services or the enterprise identity provider, ongoing burden to patch, monitor, and scale TURN servers, and dropped WebRTC connections during peak usage. Pod migrations across Availability Zones also required manual routing updates.
AWS said Kinesis Video Streams replaced the self-managed TURN layer because it offers built-in STUN and TURN infrastructure plus signaling channels, and integrates with IAM. That eliminated the need for a custom token-exchange service for time-limited, namespace-scoped TURN credentials. AWS specifically called this an important factor in reducing operational costs.
Recommendation: When evaluating infrastructure for shared media, remote collaboration, or interactive product experiences, include staffing and maintenance costs in the decision. A self-managed component may appear flexible at first, but patching, routing, and auth integration can become the hidden tax on your roadmap.
- AWS says self-managed TURN infrastructure created operational burden.
- Kinesis Video Streams removed the need for a custom token-exchange service in this setup.
- Managed services can reduce maintenance load as usage grows.
- AWS notes Kinesis Video Streams WebRTC pricing depends on signaling channels, signaling messages, and TURN streaming minutes.
What marketers and digital teams should learn from a Kubernetes security story
At first glance, this looks like a narrow infrastructure topic. But the business lesson is broader. Many modern marketing and content operations depend on shared systems: collaboration environments, creative review apps, AI-assisted production pipelines, internal demos, device testing, and customer-facing experiences. If those platforms rely on insecure defaults, the brand risk shows up later as downtime, access friction, or trust erosion.
That is especially relevant for teams using AI workflows and rapid publishing cycles. Faster output often means more internal tools, more service accounts, and more temporary environments. The AWS post underscores that speed without standardized access controls can create complexity that scales faster than the team does.
Recommendation: Document where your organization relies on shared infrastructure, who authenticates to it, and which protocols or services bypass centralized identity. Then build content and launch planning with reliability in mind. Readers can also explore PostPilot Daily’s content planning, publishing, or engagement tools as part of a broader effort to keep campaigns organized while operational systems mature.
- Infrastructure choices can affect reliability, compliance, and audience trust.
- Shared internal tools deserve the same governance mindset as public-facing channels.
- Growth teams should coordinate with platform and security teams before launching always-on shared environments.
Source and further reading
Addressing protocol security gaps in multi-tenant Kubernetes ↗AWS Security Bulletins ↗AWS Security Blog ↗Frequently asked questions
What did AWS warn EKS users about?
AWS said some protocols used in multi-tenant Kubernetes environments, including ADB and WebRTC-related components, can create security gaps because they may use unencrypted channels by default or lack built-in enterprise authentication.
What is the main risk in a multi-tenant cluster?
According to AWS, pods can reach other tenants’ open ports by default unless network policies are configured to restrict cross-namespace traffic.
What architecture did AWS describe as a stronger approach?
AWS described an architecture using namespace-scoped mTLS gateways, managed TURN relay through Kinesis Video Streams, and JWT identity chains tied to an enterprise identity provider.
Did AWS discourage any older design pattern?
Yes. AWS said not to implement the earlier bastion-based design it showed for comparison, particularly for low-latency environments, because routing through the Kubernetes API server created a throughput bottleneck.
Why does this matter outside platform engineering?
Because shared infrastructure underpins many internal and customer-facing workflows. Weak defaults in authentication, encryption, or routing can affect reliability, compliance posture, and trust across product, marketing, and operations teams.
JOIN THE CONVERSATION

No comments yet. Start the conversation.