Compliant with Local Controls, Collectively Discriminatory. A Governance Architecture for Multi-Agent AI in Regulated Finance
3.40T1 sourcearXiv cs.MA
Source record
Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2609.27994.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryA finance-specific reference architecture called ARIA addresses the gap where locally compliant AI agents can collectively produce discriminatory or unsafe outcomes. It organizes six governance capabilities across normative, execution, and assurance planes, illustrated by two simulations on credit exclusion and drift detection.
Why it mattersNames a specific failure mode (constitutional non-compositionality) where individual component checks miss collective harms, and proposes a falsifiable architecture with mapped regulatory evidence needs.
Cited by
No citations on record.
