Blog
Cross-Region Inference Needs Destination Evidence
Cross-Region inference can improve capacity and availability, but it also changes the data path. AWS has expanded access to several OpenAI GPT models in Amazon Bedrock from five Asia Pacific regions. According to AWS, global inference profiles may route requests to any supported commercial AWS Region.
The source Region is not the processing Region
An API call may start in Thailand or Singapore while the model processes it in another destination Region. The configured source Region is therefore insufficient evidence for architecture diagrams, privacy reviews, and supplier controls.
AWS states that customer data is not stored in either the source or destination Region during normal inference. The current announcement also names an important exception: traffic flagged by automated abuse detection may be retained for up to 30 days in the destination Region. Enterprises must assess both the normal path and exceptional retention rules.
Routing must become evidence
AWS records the Region actually used in the CloudTrail field inferenceRegion. This enables an auditable control path:
- Profile: Which global or geographically restricted inference profile is approved for the workload?
- Destination scope: Which possible destination Regions does that profile include at approval time?
- Evidence: Is
inferenceRegionlogged centrally, monitored, and mapped to an application? - Deviation: What happens if an unapproved Region appears or telemetry is missing?
- Exception: What retention may result from abuse detection, and who reviews it?
An automatic fallback to a global profile must not become a silent operational decision. It requires the same business, technical, and privacy approval as the primary path.
What this means for DACH enterprises
The European Commission notes that transfers of personal data outside the EU require an adequacy decision or appropriate safeguards. This does not automatically make every cross-Region route unlawful. It does make the documented processing destination part of data-flow records, transfer assessments, and processor oversight.
Do not verify only where a request starts. Define where it may be processed, how the actual destination is proven, and which exceptions trigger a new review. The Region in the SDK is configuration. The Region in the audit log is operational evidence.