Leitfaden
AI gateways vs governance control planes
Gateways standardisieren Modellzugriffe und Request-Guardrails. Governance-Control-Planes steuern Workflow-Entscheidungen mit Freigaben und exportierbaren, prüfbaren Evidence Packs.
For ML platform teams deciding between a gateway/proxy layer and a governance layer (or both) for regulated workflows.
Zuletzt aktualisiert: 17. Dez. 2025 · Version v1.0 · Keine Rechtsberatung.
TL;DR
The quick distinction
- Gateways govern requests: routing, provider abstraction, rate limits, request guardrails.
- Control planes govern decisions: approvals, overrides, escalation, and evidence about what happened.
- In regulated workflows you often need both: request safety + decision governance + exports.
Gateway
When a gateway is enough
- You mainly need centralized provider access, spend controls, and request-level guardrails.
- Your audit requirements do not include human approvals and decision evidence packs.
Control plane
When you need a control plane
- You must prove who approved/overrode a workflow decision and what context they saw.
- You need policy-as-code checkpoints that gate high-risk actions and record enforcement.
- You must export a verifiable evidence pack (manifest + checksums) for auditors.
Referenzen