Causal Evidentiary Governance for High-Risk Machine Learning Systems
A research paper introduces Causal Evidentiary Governance (CEG), a framework for regulated institutions to commit to a versioned directed acyclic graph (DAG) partitioning causal pathways into allowable and disallowed groups. The Causal Harm Rate measures prediction variation attributable to disallowed causal pathways. Each decision is accompanied by a signed Decision-Evidence Packet (DEP), cryptographically binding the prediction to a digest of the published DAG and path-specific attributions. DEP digests can be appended to a Merkle tree to enable logarithmic-cost inclusion proofs. The framework is validated using demographic summaries from four years of PMA credit supervisory data to construct 10,000 synthetic credit applicants across four strategic DAG counterfactuals.
The paper proposes Causal Evidentiary Governance (CEG) to address limitations of current fairness governance practices under regulations like the EU AI Act and GDPR. CEG uses a versioned DAG to separate causal pathways, computes a Causal Harm Rate, and issues signed Decision-Evidence Packets with Merkle tree inclusion proofs for efficient verification. Validation uses synthetic credit applicant data derived from PMA supervisory data.
CEG combines causal DAG versioning, path-specific attribution, and cryptographic commitments (signed DEPs and Merkle trees) to enable efficient, verifiable governance of high-risk ML decisions. The Causal Harm Rate provides a quantitative measure of disallowed pathway influence, and Merkle tree inclusion proofs offer logarithmic verification cost.
This framework targets regulated sectors such as credit, hiring, and resource distribution, where compliance with the EU AI Act and GDPR requires auditable fairness. It may influence how institutions design ML governance and audit systems, potentially reducing compliance overhead through efficient evidentiary verification.
CEG could lower the cost and complexity of regulatory compliance for high-risk ML systems by providing a standardized, verifiable method for causal fairness governance. It may also reduce legal and reputational risks associated with biased automated decisions.
Observable next signals include adoption of CEG-like frameworks in regulatory guidance, pilot implementations by financial institutions, and further research on causal harm metrics and cryptographic audit trails. The paper's validation on synthetic data suggests a need for real-world deployment studies.