1 Why This Matters
AI and algorithmic systems are increasingly deployed in high-stakes environments—financial markets, healthcare, autonomous vehicles, and critical infrastructure. These systems often operate as "black boxes," making decisions at speeds and scales that exceed human comprehension.
Traditional audit mechanisms rely on mutable logs and trust-based assertions. Such approaches are insufficient when:
- Decision-making processes are opaque and difficult to reconstruct
- Log integrity cannot be independently verified
- Regulatory bodies require evidence, not mere assertions
Regulatory frameworks—including the EU AI Act (Article 12: automatic recording of events for high-risk AI systems), MiFID II record-keeping requirements, and SEC record-preservation rules (Rule 17a-4)—require records to be kept, and some attach integrity conditions: Article 72(1) of Delegated Regulation (EU) 2017/565 requires that amendments be ascertainable and that records cannot otherwise be manipulated or altered, and SEC Rule 17a-4(f) requires either a complete time-stamped audit trail or a non-rewriteable, non-erasable format. Article 12 of the EU AI Act sets no integrity standard for the logs. None of them prescribes a cryptographic method. VSO's position is that verifiable, tamper-evident audit trails make such records independently checkable.
Core Principle: "Verify, Don't Trust"—audit evidence should be cryptographically verifiable by independent third parties, without reliance on the system operator's assertions.
2 What is VSO
VeritasChain Standards Organization (VSO), operated by VeritasChain Co., Ltd., develops open specifications for verifiable audit evidence in AI and algorithmic systems. Its neutrality is described by public procedures, conflict-of-interest management, and a non-endorsement policy.
VSO operates under the following principles:
- Neutral by mechanism: specifications are published under CC BY 4.0, conformity assessment is assigned to VSO-approved Conformity Assessment Bodies rather than to VSO (the only CAB approved so far is VeritasChain Co., Ltd., which also provides VSO's operating base and is therefore not independent of the scheme owner), and the non-endorsement policy is public. The operating base is provided by VeritasChain Co., Ltd.
- Vendor-neutral: Specifications are designed to be implementation-agnostic
- Open development: Technical work is conducted transparently with public documentation
Important Clarifications:
- • VSO does not endorse or recommend specific products or vendors
- • VSO does not grant regulatory approval or certification of regulatory compliance
- • VSO does not provide investment, legal, or financial advice
3 What is VCP
VeritasChain Protocol (VCP) (current version: v1.2.0 General Availability; tag v1.2.0, commit b23acb266f88df81d960d36d58fb64ceaa0dd5ed; GA merged 2026-10-08 JST; fix merged 2026-10-09 JST; separate from VAP v1.2.0 Draft 3) is an open technical specification designed to enable cryptographically verifiable audit evidence for algorithmic and AI-driven systems.
VCP is intended to function as a "flight recorder" for algorithmic decisions—capturing not just outcomes, but the context and parameters that led to those outcomes, in a format that can be independently verified.
Core Technical Components
- Merkle trees and external anchoring: Event hashes are collected in Merkle trees (RFC 6962 domain-separated hashing) whose roots are anchored externally at tier-dependent intervals (required at every tier since v1.1); hash chaining between events (PrevHash) is optional
- Merkle proofs: Efficient verification of individual records within large datasets
- Digital signatures: Each event is signed to establish authenticity and non-repudiation
- Tamper-evident event logs: Any modification to historical records becomes detectable
- Third-party verifiability: Audit evidence can be verified without access to the original system
Integration Approach
VCP is designed for sidecar integration—it aims to capture audit evidence without requiring invasive modifications to existing trading or decision-making systems. This approach is intended to minimize integration complexity and operational risk.
4 What Makes It Different
| Aspect | Traditional Logging | VCP Approach |
|---|---|---|
| Integrity Model | Trust-based (assumes logs are unmodified) | Cryptographically verifiable (tampering is detectable) |
| Verification | Requires access to internal systems | Third-party verifiable without system access |
| Evidence Type | Assertions by system operator | Cryptographic proofs |
| Audit Scope | Typically outcomes only | Decision context, parameters, and outcomes |
| Modification Detection | Difficult or impossible | Detectable through Merkle proofs and external anchors (and through hash-chain breaks where the optional PrevHash is used) |
5 Current Status & Public Engagement
VSO is engaged in ongoing efforts to advance the development and adoption of verifiable audit standards:
- Technical documentation: VCP v1.2.0 specification (General Availability: tag v1.2.0, commit b23acb266f88df81d960d36d58fb64ceaa0dd5ed; GA merged 2026-10-08 JST; fix merged 2026-10-09 JST; separate from VAP v1.2.0 Draft 3), JSON Schema, the VAP v1.2.0 Draft 3 metaframework, and supporting materials are publicly available
- Standards body engagement: five individual Internet-Drafts are active at the IETF (none adopted by any Working Group); VSO's liaison statement to ITU-T SG17 was returned by the TSB; no ISO/IEC submission has been made
- Research publications: Position papers and technical analyses have been published for public review
- Open tools: Reference implementations and evaluation tools (such as the VAP Scorecard Explorer) are available for public use
- Outreach: VSO has sent its specifications and technical notes to regulatory authorities and standards bodies in several jurisdictions; a submission does not imply any response, endorsement or approval
6 Relationship to VAP Framework
VCP operates within the broader context of the Verifiable AI Provenance (VAP) Framework:
- VAP Framework (v1.2.0 Draft 3): A high-level metaframework defining general requirements for verifiable AI decision provenance across multiple domains
- VCP: A domain-specific profile of VAP, initially focused on financial services and algorithmic trading
This modular architecture is designed to allow future extensions to other high-risk domains—such as autonomous vehicles, healthcare AI, and critical infrastructure—while maintaining a consistent approach to verifiable audit evidence.
Note: Domain-specific profiles beyond VCP include CAP v1.0 and CPP v1.4 (Released); OAP and MAP (Working Drafts); PAP (Partial Draft); IAP (Announced); and DVP, AAP and EIP (Planned). DAP and SMP are cross-cutting capabilities, not domain profiles (both Working Drafts). An application example that combines OAP and DAP is described in Verifiable Evidence for AI-Assisted Security Research (design proposal, non-normative).
7 Important Clarifications
To ensure accurate understanding of VSO and VCP, the following points are explicitly stated:
VCP is a voluntary specification: it has no legal or regulatory force, and it has not been adopted by any recognized standards body or regulatory authority.
VCP does not constitute regulatory guidance. Compliance with applicable laws and regulations remains the responsibility of implementing organizations.
Implementation of VCP does not imply compliance or certification by default. Formal compliance assessment requires separate evaluation.
Use of VCP does not guarantee regulatory acceptance. Regulatory outcomes depend on jurisdiction-specific requirements and interpretations.
For details on VSO's organizational design and governance principles, see our Distributed Organization policy.