arXiv: Split Tallies: A Discrete Certificate Calculus for Auditing Dynamic Ordered Sets in Constant Memory

This publication introduces a novel cryptographic method called Split Tallies, which enables the auditing of dynamic ordered sets, such as transaction logs or supply chain records, using only constant memory. The paper presents a discrete certificate calculus that allows verifiers to confirm the integrity and ordering of data entries without storing the entire dataset. This is a technical proposal from the academic computer science community, not a binding regulatory text, but it directly addresses the growing need for efficient, privacy-preserving audit mechanisms under the AI Safety framework.

Organizations that manage large, continuously updated ordered datasets will be most affected. This includes financial institutions handling transaction ledgers, logistics firms tracking shipments, and any AI system that relies on verifiable data provenance. Regulated entities subject to audit requirements under the EU AI Act or similar frameworks should monitor this development, as it offers a potential technical pathway to meet transparency and accountability obligations without prohibitive storage costs.

Compliance teams should first assess whether their current audit processes for ordered data sets are memory-intensive or prone to scalability issues. Next, engage with technical teams to evaluate the feasibility of implementing Split Tallies in existing systems, particularly for AI model training logs or inference records. Finally, prepare to brief legal and risk officers on how this method could support future regulatory audits, while noting that it remains an academic proposal not yet endorsed by any regulatory body.