Sovereign AI for Supply Chain Due Diligence: A Standardized Compliance Framework for the German LkSG

Background & Context The German Supply Chain Due Diligence Act (Lieferkettensorgfaltspflichtengesetz, LkSG), in force since January 2023, obligates companies with 1,000 or more employees to identify, prevent, and mitigate human rights and environmental risks across their supply chains.

The Problem Despite this clear regulatory mandate, the industry's approach to compliance remains fragmented:

• Suppliers report data in inconsistent formats.

• Risk scores are calculated through manual and unreproducible processes.

• Dominant ESG platforms create data sovereignty risks through centralised, cloud-based architectures.

Methodology & Solution This thesis addresses these challenges by designing, building, and validating a sovereign AI-driven compliance framework tailored for the renewable energy and green hydrogen sector. Following the Design Science Research (DSR) methodology (Peffers et al., 2007), the proposed framework operates as a four-layer headless microservice:

1. Standardised JSON Schema: Extends the WBCSD PACT protocol with LkSG §4–§10 compliance data.

2. Deterministic Rule Engine: Produces transparent, reproducible risk scores.

3. Sovereign AI Narrative Layer: Utilizes Mistral 7B running entirely offline via Ollama to ensure data privacy.

4. FastAPI Audit Trail: Generates immutable, legally defensible assessment artifacts.

To ensure reliability, a hallucination guard and human-in-the-loop mechanism guarantee that the AI remains assistive while human compliance officers retain ultimate decision-making authority.

Validation & Impact The framework was validated through rigorous technical testing across three distinct supplier risk profiles. It was further refined through structured consultations with eight industry practitioners from DHL, DHL Supply Chain, DHL IT Services, and other key organisations.

A stakeholder impact analysis demonstrates significant value creation across multiple dimensions:

• Economic: Compliance cost reduction.

• Political: Enhanced data sovereignty.

• Social: Improved worker protection.

• Environmental: Enforcement of Green Hydrogen certifications.

Key Contributions This thesis contributes the first open human rights due diligence (HRDD) data model extension for the WBCSD PACT ecosystem. Ultimately, it offers a sovereign, transparent, and extensible alternative to traditional, centralised ESG compliance platforms.

Keywords: LkSG, supply chain due diligence, sovereign AI, WBCSD PACT, JSON Schema, Green Hydrogen, compliance automation, human-in-the-loop, Design Science Research