marcomb.github.io

Sovereign AI != Compliant AI

Data residency solves where data lives. The EU AI Act regulates what the system does with it.

What You Built: Data Sovereignty

  • Data residency: control where data is stored and processed.
  • Self-hosted infrastructure: deploy on your own servers.
  • GDPR compliance: data protection by design and by default.
  • Encryption and access control: protect data at rest and in transit.

What You Still Need: AI Act Compliance

  • Article 6 classification (system use case): is the deployed AI system high-risk given its intended purpose and deployment context (Annex III)?
  • GPAI model due diligence (Article 51/55 implications): if the solution relies on a frontier GPAI model, require model identification (name/version), model-change notifications, and evidence of upstream risk controls from the supplier; treat this as a procurement and governance requirement, separate from Annex III classification.
  • Article 9 risk management: documented, continuous, auditable risk assessment and mitigation across the lifecycle.
  • Article 12 automatic logging: system-level logging for traceability and incident investigation, not only platform metrics.
  • Article 14 human oversight: who intervenes, how, and when, including escalation paths and stop/override mechanisms.
  • Article 26 deployer obligations: your customers’ operational responsibilities (configuration, monitoring, training, incident handling).
  • Shared responsibility framework: where your obligations end and where theirs begin, contractually and operationally.

Key Point

Data sovereignty is an infrastructure decision. AI Act compliance is a risk-and-governance architecture. One does not automatically deliver the other.