Strict Liability Shields: Algorithmic Provenance as a Compliance Mechanism for the EU AI Act

Abstract

The implementation of the EU AI Act (Regulation (EU) 2024/1689) introduces a "Strict Liability" regime for High-Risk AI systems, specifically under Article 10 (Data Governance) and Article 12 (Record Keeping). This paper argues that current industrial data governance practices—reliant on probabilistic "Data Quality" and passive observability—are legally insufficient to meet these new enforcement standards.

We demonstrate that "Semantic Drift" in upstream data pipelines constitutes a foreseeable failure mode that triggers liability during inference. In response, we propose the Open Data Governance Standard (ODGS), a deterministic protocol that replaces passive monitoring with active liability shielding. By introducing a cryptographic "Hard Stop" mechanism at runtime, the protocol ensures that no inference occurs on unverified data, effectively translating the abstract legal requirement of "error-free data" into an executable technical constraint. This framework provides a reference implementation for Risk Officers and General Counsel seeking to bridge the gap between regulatory text and engineering reality.

Keywords: EU AI Act, Article 10, Strict Liability, Algorithmic Governance, ODGS, Compliance, Risk Management.

JEL Classification: K13, K20, O33, C80