AI connectivity is becoming the control plane for agentic traffic

At a glance

What this is: This is Kong’s view that AI connectivity is becoming the control plane for agentic traffic, with tokens and agent-to-agent exchanges becoming the new governance surface.

Why it matters: IAM, NHI, and platform teams need to treat AI traffic as a runtime access problem, because routing, permissions, auditability, and blast-radius control are converging in the same layer.

Context

Kong’s article is about the governance gap that appears when AI systems stop behaving like single applications and start moving tokens, context, and actions between models, agents, and tools. In that environment, traditional API management is no longer the whole answer because the traffic itself carries intelligence and access decisions.

The identity problem is not just who authenticates to what. It is how machine-to-machine flows are authorised, routed, observed, and constrained once agents begin consuming services at scale and in real time. For IAM practitioners, this is a control-plane problem that spans NHI, workload identity, and policy enforcement.

The article’s core claim is that the new connectivity layer sits between AI agents and enterprise systems, and that this layer will need the same kind of governance discipline that API gateways once brought to microservices. That is an atypical but increasingly plausible pattern for organisations moving from pilot AI usage to production agentic workflows.

Key questions

Q: How should security teams govern AI agent traffic at the control plane level?

A: Security teams should govern AI agent traffic where routing, authentication, and auditability intersect, not only where applications connect. The right model treats the traffic layer as a policy enforcement point for tokens, prompts, tool calls, and delegated access, so machine-paced workflows remain observable and containable across systems.

Q: Why does agentic AI change the way access and routing should be controlled?

A: Agentic AI changes access and routing because machine workflows can chain actions faster than human review cycles can follow. When a session can discover, learn, and transact across multiple systems, the control question becomes how far a permitted action can spread, not just whether the first request was allowed.

Q: What breaks when AI gateway controls are treated like ordinary API security?

A: Ordinary API security assumes stable clients and predictable request paths. Autonomous agents can switch tools, chain calls, and change execution intent mid-session, so the gateway must enforce identity context and policy continuity across multiple steps. Without that, the organisation sees traffic but not the behavioural shift that creates risk.

Q: How do teams reduce the blast radius of agentic workflows?

A: Limit the agent's ability to move from observation to mutation in one path. Use narrow credentials, separate approval boundaries, and human review before high-risk sinks such as commits, deployments, or database writes. The goal is to stop the workflow before tainted context turns into operational change.

Technical breakdown

Why token traffic becomes the control surface

Tokens are not just model inputs and outputs. In agentic workflows they carry context, memory, and sometimes delegated intent across multiple services and protocols, which makes them part of the security boundary rather than a simple payload. Once agents consume tokens directly from other agents, the traffic layer becomes the only practical place to inspect, route, and constrain those exchanges consistently. That shifts governance away from the application edge and toward the proxy or gateway layer, where policy can follow the traffic instead of being rebuilt in each backend integration.

Practical implication: Treat token flow as governed access traffic, not just application data, and place controls where the exchange is actually happening.

Why AI gateways inherit API gateway responsibilities

The article maps the cloud-native API gateway pattern onto agentic AI, where connectivity logic moves out of individual services and into a shared broker layer. In practice that means authentication, routing, rate limiting, logging, and policy enforcement converge around AI-specific traffic, including MCP sessions and model calls. The architectural shift matters because it turns a fragmented set of point controls into a single control plane, but it also means the gateway now influences both security posture and operational behaviour. The more agentic the environment becomes, the more the gateway functions as runtime governance rather than simple transport plumbing.

Practical implication: Design gateway policy for AI traffic as a runtime governance layer, not as an add-on to ordinary API management.

How hypervolume changes governance economics

Kong frames the AI era around hypervolumes, with traffic growth moving from billions to trillions and beyond. At that scale, performance and governance are no longer separate concerns because latency, token budgets, auditability, and blast radius all affect whether the system is usable and safe. The important architectural point is that control decisions must be cheap enough to run at machine scale. If they are not, teams will either bypass them or push them deeper into the stack, where visibility and consistency degrade.

Practical implication: Assume AI traffic controls must be low-latency and policy-driven, or they will be bypassed under production pressure.

NHI Mgmt Group analysis

AI connectivity is becoming an identity control plane, not just a transport layer. Once agents exchange tokens, prompts, and tool calls directly, the network path starts carrying authorisation context as well as data. That means the control point shifts from static integration design to runtime governance over non-human traffic. Practitioners should read this as a change in where identity policy must be enforced.

The assumption that API security alone can govern agentic traffic is already collapsing. API controls were built for request-response systems with relatively stable consumers, not for autonomous or semi-autonomous machine traffic that discovers, chains, and reuses capabilities in-session. The implication is not merely more controls, but a different governance model for machine intent, scope, and auditability.

Blast-radius control becomes the defining NHI governance concept in agentic systems. When a single workflow can fan out across models, tools, and downstream APIs at machine speed, the security question is no longer only access granted, but how far a permitted action can propagate. That makes containment, policy propagation, and traffic segmentation central to NHI oversight.

Traffic-layer governance will increasingly determine whether agentic AI can be operated safely at scale. The article’s architecture points to a future where routing, rate limiting, observability, and access control are inseparable from AI delivery. For identity leaders, that means the governance boundary is moving closer to runtime orchestration and away from siloed app-by-app review.

Machine-to-machine consumption changes how accountability works in practice. Human reviewers cannot reliably keep pace with token-level decisioning when agents transact faster than operators can observe. The field needs governance patterns that account for machine-paced execution, because audit trails and access reviews that assume human cadence will miss the operational reality.

What this signals

AI connectivity will push identity governance closer to runtime orchestration. As agents begin to transact across multiple systems in a single workflow, access review alone is no longer enough because the meaningful decision happens during execution. Programmes that still centre only on application onboarding or post-hoc certification will struggle to keep pace with machine-speed traffic.

The control plane for agentic systems will blur the boundary between IAM and network governance. Routing, policy enforcement, observability, and authorisation are converging in the same operational layer, which means identity leaders need to work more closely with platform and API teams. The practical shift is toward governed machine flow, not isolated access decisions.

Token-aware governance will become a core design requirement for AI operations. If tokens are the unit carrying context and delegated action, then permissioning, logging, and containment all need to understand that unit of exchange. Teams that can trace and constrain token movement will be better placed to manage agentic risk without slowing production unnecessarily.

For practitioners

• Map AI traffic to a named control plane Identify where model calls, MCP sessions, and agent-to-agent exchanges currently bypass policy enforcement, then decide which layer owns authentication, routing, and audit logging.
• Segment agentic flows by blast radius Classify agent workflows by the downstream systems they can reach, and apply tighter routing and permission boundaries to high-impact paths such as finance, customer data, and admin tools.
• Make token governance part of access design Treat token issuance, propagation, and reuse as part of the access model so that delegated machine activity can be constrained at runtime instead of only at integration time.
• Instrument auditability at the traffic layer Log the identity, route, tool, and policy decision for each agentic exchange so incident responders can reconstruct what a machine flow did without chasing separate application logs.
• Stress-test controls for hypervolume Validate that gateway policy, inspection, and observability still work when AI traffic scales far beyond normal API workloads, because controls that fail under load will not hold in production.

Key takeaways

• Agentic AI moves governance from human-facing application access toward the traffic layer where tokens, routing, and runtime policy converge.
• The article’s core message is that AI connectivity is becoming a control plane problem, not just an API management problem.
• For practitioners, the priority is to govern machine-paced flows with containment, auditability, and low-latency policy enforcement.

Key terms

• AI Connectivity: AI connectivity is the enterprise control layer for governing how applications, users, and agents reach AI services. It combines routing, security, observability, and policy enforcement so AI interactions can be managed consistently across providers and use cases.
• Token Traffic: Token traffic is the flow of machine-readable units that carry context between models, agents, and tools. For agentic systems, tokens are not just payloads; they are the operating medium through which access, reasoning, and downstream actions are expressed.
• Control Plane: The control plane is the set of actions that create, configure, or manage a service. For AI workloads, it covers deployment and administration of the model platform, while data-plane permissions govern what the service and its identities can read or process.
• Blast Radius: The potential scope of damage if a specific credential or identity is compromised. Identities with broad permissions have a larger blast radius and represent a higher priority for least-privilege enforcement and security controls.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.