Right to Local Intelligence manifesto calls for mandatory on-device AI inference outside EU AI Act scope

The site defines local intelligence as inference confined to edge silicon with no outbound model weights or activations. It contrasts this with current cloud-dependent systems from OpenAI and Google, where query logs remain accessible under provider policies. Hacker News discussion reached 71 points with 30 comments focused on implementation feasibility rather than the sovereignty claim itself.

No quantitative benchmarks appear in the source. Related deployments include Apple's Private Cloud Compute attestation protocol and Qualcomm's 2024 Snapdragon X Elite NPU delivering 45 TOPS locally. EU AI Act Article 6 classifies high-risk systems by use case but imposes no location-of-compute restrictions, leaving the proposal unaddressed in existing text.

Operational impact centers on hardware certification requirements. Device vendors would need to publish verifiable execution environments, similar to TPM 2.0 attestation chains. Regulators currently track data flows under GDPR Article 44; shifting enforcement to silicon-level isolation would require new audit standards absent from the AI Act's conformity assessments.

Next steps hinge on national implementation. At least two EU member states have signaled interest in edge mandates during 2025 consultations, independent of Brussels timelines.