Constant-Size Cryptographic Evidence Structures for Regulated AI Workflows

TL;DR

The paper tackles the challenge of provable auditability in regulated AI by proposing constant-size cryptographic evidence structures that uniformly encode per-event audit data. It introduces a formal model, a fixed-size evidence syntax with $k$ $\lambda$-bit fields, and a generic hash-and-sign construction that binds events to their inputs, outputs, and environment, while enabling seamless integration with hash-chain and Merkle-based audit frameworks and optional TEEs. The authors provide informal security arguments for event binding, tamper detection, and chain integrity, analyze per-event costs $\Theta(k)$ hash evaluations plus a constant number of signature operations, and validate scalability with a prototype and microbenchmarks showing predictable, hardware-friendly performance. The work aims to standardize a compact, verifiable audit representation suitable for clinical trials, pharmaceutical compliance, and medical AI governance, with practical guidelines on field allocation, encoding design, and anchoring frequency. Overall, it offers a rigorous abstraction that addresses metadata leakage, verification cost variability, and integration challenges in regulated AI audit trails, while outlining avenues for formal security definitions and broader standardization.

Abstract

This paper introduces constant-size cryptographic evidence structures, a general abstraction for representing verifiable audit evidence for AI workflows in regulated environments. Each evidence item is a fixed-size tuple of cryptographic fields, designed to (i) provide strong binding to workflow events and configurations, (ii) support constant-size storage and uniform verification cost per event, and (iii) compose cleanly with hash-chain and Merkle-based audit constructions. We formalize a simple model of regulated AI workflows, define syntax and algorithms for evidence structures, and articulate security goals such as audit integrity and non-equivocation. We present a generic hash-and-sign construction that instantiates this abstraction using a collision-resistant hash function and a standard digital signature scheme. We then show how to integrate the construction with hash-chained logs, Merkle-tree anchoring, and optionally trusted execution environments, and we analyze the asymptotic complexity of evidence generation and verification. Finally, we implement a prototype library and report microbenchmark results on commodity hardware, demonstrating that the per-event overhead of constant-size evidence is small and predictable. The design is informed by industrial experience with regulated AI systems at Codebat Technologies Inc., while the paper focuses on the abstraction, algorithms, and their security and performance characteristics, with implications for clinical trial management, pharmaceutical compliance, and medical AI governance.

Figures (1)

• Figure 1: Overview of the evidence workflow from workflow events to immutable storage and verification.