API gateway

Governance

The single controlled entry point to a system, where every request is authenticated and inspected before it reaches any service.

What is an API gateway?

An API gateway is the single entry point through which external requests reach a system. It authenticates each request, checks it against defined rules, applies rate limits and passes permitted traffic to the right internal service. Many deployments place a web application firewall in front of or within the gateway to block known attack patterns before any application code is reached.

Also called: Secure gateway, Edge gateway

Why one entry point is easier to defend than many

Every service that accepts traffic directly from outside is another place where authentication and rate limiting must be implemented correctly. One mistake in any of them exposes the whole system. A gateway concentrates those controls in a single component that can be tested and monitored as one. Internal services then accept traffic only from the gateway, which turns a scattered perimeter into a defined one.

  • Authentication is enforced once instead of in every service separately.
  • Rate limits protect internal services from overload and abuse.
  • A single access log records every request that entered the system.
  • Internal services can refuse any traffic that bypassed the gateway.

How an API gateway works

  • All external traffic is routed to the gateway over encrypted connections.
  • Each request is authenticated against the organisation's identity provider.
  • Requests are inspected against firewall rules and expected formats.
  • Rate and size limits are applied per client and per route.
  • Permitted requests are forwarded to the internal service for that route.
  • Every request and decision is logged and sent to security monitoring.

How an API gateway is evaluated

An API gateway is evaluated by testing what it admits and what it refuses. Check the following through regular testing:

  • requests admitted without valid authentication
  • internal services reachable without passing through the gateway
  • rate limits enforced under deliberate load
  • malformed and hostile requests blocked at the edge
  • latency added by the gateway at peak traffic

Limitations, risks and common failure modes

  • A route added for testing bypasses the gateway and is never removed.
  • Rate limits set too generously offer no real protection.
  • The gateway becomes a single point of failure without redundancy.
  • Logs are collected but never reviewed by anyone.

Related terms

Deny-by-default egressPermitting outbound connections only to destinations explicitly allowed, and refusing everything else.Tenancy isolationKeeping one organisation's data and compute separate from every other organisation's.Observability for AI systemsInstrumentation that records what happened inside a request, not only whether it succeeded.Prompt injectionInstructions hidden in content a model reads, intended to redirect its behaviour.Red teamingAdversarial testing before deployment, carried out by people trying to make the system fail.Service accountA non-human identity with its own entitlements, used so that automated actions are attributable.

All 315 terms