CSE Roadmap

Specialist tier · about 3 hours · 12 parts

Part 4 of 12 · 1 min read

A DPIA sits inside the obligation table, but it comes from a different law

The AI Act's high-risk tier and GDPR's rules on personal data overlap in practice but are separate regimes with separate triggers, and a DPIA is where that overlap shows up as one obligation entry.

Think of two building codes that both apply to the same hospital, such as a fire code and an accessibility code. Each is inspected separately. A single ramp might still satisfy a requirement in both.

The AI Act's tiers and GDPR's rules on personal data are two different laws that often apply to the same agent at once. A DPIA is simply where GDPR's own obligation shows up inside this module's one-tier-one-list simplification.

A worked example. In the table above, dpia is required only for the high tier, alongside human_oversight and decision_logging. That is this module's AI-Act-shaped simplification of a GDPR requirement that actually has its own trigger: high-risk processing of personal data, which a credit-scoring agent like loan-triage almost certainly has regardless of which AI-Act tier it lands in.

The key lines. From code/mini-compliance/compliance.py:

# from code/mini-compliance/compliance.py
OBLIGATIONS = {
    "dpia": {"high"},
    "human_oversight": {"high"},
    "decision_logging": {"high"},
}

Read it as: dpia is just one more key in the same dictionary as the AI-Act-specific obligations. The code does not distinguish which law an item comes from. That is exactly the simplification the module's "Not real" box warns about.

GDPR, DPIA, Article 22

  • GDPR: the EU's data protection regulation.
  • DPIA: data protection impact assessment, required for high-risk processing of personal data.
  • Article 22: limits on decisions based solely on automated processing with legal or similarly significant effects.

Example: loan-triage denying credit automatically is exactly the kind of "legal or similarly significant effect" Article 22 limits, independent of whether the AI Act also calls it high-risk.