The rapid growth of edge-to-cloud applications has intensified the need to guarantee that sensitive data remains within trusted jurisdictions while still meeting performance targets. Existing orchestration stacks—Kubernetes for compute and software-defined networking (SDN) controllers such as ONOS for traffic steering—expose rich policy knobs, yet they still require experts to translate high-level privacy goals into dozens of low-level rules. This paper proposes and experimentally validates an LLM-driven, intent-based framework that enforces privacy requirements across both computing and networking layers of the cloud–edge continuum. A GPT-4o model interprets natural-language privacy intents, then automatically (i) generates Kubernetes node-selector rules to place pods only on nodes that satisfy privacy and locality constraints, and (ii) compiles SDN flow rules that confine packets to compliant paths. To benchmark the approach, we curated a corpus of 90 diverse privacy intents spanning computing, networking, and hybrid scenarios and built an automated validator that integrates a Kubernetes + ONOS test-bed, deploys model-generated configurations, and issues pass–fail reports without human intervention. Experiments show that GPT-4o converts intents into correct, enforcement-ready policies in 95.6% of 90 end-to-end trials while holding average latency to around 21 seconds. The unified validator further demonstrates that continuous, cross-layer compliance checking can be executed in seconds, eliminating hours of manual inspection and reducing error rates.