The Flight Recorder for AI
Aircraft have flight data recorders. Nuclear plants have detailed monitoring systems.
But AI systems making millions of decisions per second? Almost none have tamper-evident records.
VAP addresses part of this problem.
"Encoding Trust in the AI Age"
Official VAP Specification
github.com/veritaschain/vap-spec
Maintained by VeritasChain Standards Organization (VSO)
The regulatory landscape and incident trends demand a new approach
Tamper Detection
Merkle tree + external anchoring
Causal Traceability
Input → Decision → Outcome
Legal Evidence
Verifiable cryptographic evidence
A cross-domain meta-framework for cryptographically verifiable AI decision provenance
VAP (Verifiable AI Provenance Framework) specifies the requirements for cryptographically verifiable decision provenance common to all high-risk AI systems — a cross-domain meta-framework.
VAP is NOT "regulation that stops AI."
VAP IS "provenance infrastructure that enables AI to continue operating safely."
VAP is NOT an implementation specification.
Actual implementations are domain-specific profiles: VCP (Finance), MAP (Medical), DVP (Automotive), EIP (Energy), PAP (Public Policy), IAP (Industry), CAP (Content / Creative), CPP (Capture). Status per the VAP v1.2 registry (§4.5.2): VCP v1.2 Released; CAP v1.0 and CPP v1.4 Released; OAP v0.1.1 and MAP v0.1.2 Working Drafts; PAP Partial Draft (Track B v0.2.0); IAP Announced; DVP, AAP and EIP Planned. No specification document exists yet for IAP, DVP, AAP or EIP.
The VAP family also includes the proposed VAP-LAP (Legal AI Profile). Explore individual Internet-Draft -05 and the AILEX product reference.
Application example (design proposal, non-normative; not a profile): AI security research: verifiable audit trails and evidence design — combining the OAP profile with the DAP cross-cutting capability.
Limits: What VAP does not do, set out by the patterns that recur in VSO’s AI incident verification records.
Implementation status
As of 25 October 2026: zero external implementations of VAP or any VAP profile, and zero Evidence Packs accepted in any proceeding. The previously published paid-service contract figure was withdrawn on 24 October 2026 pending reconciliation with primary records; this does not establish a count of zero.
Working Draft texts are not yet public: the texts of OAP, MAP and PAP (Track B), and of the DAP and SMP capabilities, have not been published in any VSO repository. The registry gives each document's status, not its availability.
VAP defines "the minimum conditions that must be met."
Think of VAP as the "interface contract" — domain profiles are the "implementations."
VAP targets domains where: "When it fails, people or society seriously die or collapse."
Systems where failures cause irreversible damage to human life, social infrastructure, or democratic institutions.
Domains where the provenance of AI decisions matters
VeritasChain Protocol
Algorithmic trading audit trails, HFT systems, AI-driven trading strategies
Medical AI Profile
AI diagnostic systems, imaging analysis, treatment recommendations
Driving Vehicle Profile
Autonomous driving (L3-5), ADAS, aviation AI, drone control
Energy Infrastructure Profile
Smart grid AI, power network management, critical infrastructure
Public Administration Profile
Credit scoring, welfare decisions, immigration AI, hiring algorithms
Industry Accountability Profile
Sector-specific extensions, industry-tailored governance, regulatory compliance
Content / Creative AI Profile
Games, film, animation, publishing, music — cryptographic AI usage evidence
Capture Provenance Profile
Media capture verification — make verifiable when, where, and by whom photos/videos were taken
Legal AI Profile
Legal queries, drafting, fact checking and human review — connect legal AI work with inspectable records.
Proposed profileAcross all domains, AI transparency and traceability is not "nice to have" — it's "civilization cannot function without it."
VAP / VSO / Domain Profiles — A three-layer standardization structure
Conceptual Meta-Framework
Defines the minimum requirements common to all domains — the abstract layer for AI decision provenance
defines & maintains
Standards Organization
The organization that develops and maintains VAP — maintains consistency across profiles (it does not issue certifications)
publishes profiles
Finance
v1.2 ReleasedMedical
Automotive
Energy
Public
Industry
Content / Creative
Capture
Legal AI
Proposed profileDomain profiles and proposals
Technical building blocks. The VAP v1.2 architecture (§3.1) has five layers: Integrity, Provenance, Traceability, Accountability and Domain Profiles
Integrity Layer
Decision Origins
Time Consistency
Future-Proofing
All VAP domain profiles (VCP, MAP, DVP, EIP, PAP, IAP, CAP, CPP) share a common cryptographic foundation
The proposed VAP-LAP explores these mechanisms for legal AI; its mapping is described in individual Internet-Draft -05.
One Core, Many Profiles — Domain profiles extend VAP but never replace the Shared Assurance Core.
JCS (RFC 8785) — Deterministic JSON serialization for consistent hashing across implementations.
UUIDv7 — Time-ordered unique identifiers enabling temporal ordering and global uniqueness.
SHA-256 (default) / SHA3-256 — Merkle inclusion proofs and/or per-event PrevHash chaining make event order and membership verifiable (INT-004). PrevHash is optional at framework level (INT-004a).
Merkle Trees — Batch verification and external anchoring of signed Merkle roots (TSA, transparency log or public blockchain) — required at every conformance level in v1.2 (INT-006). Each anchor record binds the event count, the first and last event IDs and the policy ID (Completeness Invariant, INT-008).
Ed25519 (default) + ML-DSA (FIPS 204, EXPERIMENTAL) — Classical signatures with a defined key lifecycle; experimental post-quantum signatures for hybrid deployments.
Standardized Verify Procedure — Defined proof structure and verification algorithm for cross-domain interoperability.
Finance
Medical
DVPDriving
EIPEnergy
PAPPublic
IAPIndustry
CAPContent / Creative
CPPCapture
VAP-LAPLegal AI
Proposed profileThe Shared Assurance Core defines VAP’s common requirements. Each profile documents its status and mapping.
Explanation ≠ Verification — Understanding the fundamental difference
| Aspect | Explainable AI (XAI) | Verifiable AI (VAP) |
|---|---|---|
| Question Answered | Why was this decision made? | Can we verify this decision actually happened? |
| Output Type | Post-hoc interpretation | Cryptographic evidence |
| Tamper Evidence | None | Detectable via Merkle proofs and external anchors |
| Analogy | "Let me explain with a PowerPoint" | "Here's the black box data" |
| Evidence type | Explanation; not independently verifiable | Verifiable cryptographic evidence; its weight is for courts and authorities to decide |
XAI answers "Why?" — VAP answers "Did it really happen, and can you verify it?"
VAP is designed to produce evidence relevant to current and emerging international regulations
Automatic logging for high-risk AI
Algorithmic trading recording requirements
Data protection & right to erasure (Crypto-shredding)
Consolidated Audit Trail
Critical infrastructure security
Medical AI device guidance
Path toward international recognition and adoption
draft-kamimura-vap-framework-02 (active individual submission) — not adopted by any IETF Working Group; no IETF standing
Engagement with ISO Technical Committee for financial services standardization
Harmonization with international AI standards committee
Potential IEEE standardization track
VSO develops and maintains VAP as a standards organization for AI decision provenance
VSO works at the same layer as W3C, IETF, IEEE, or FIX — it defines rules and does not itself certify. This is a structural comparison, not a claim of equivalent status, recognition, or adoption.
Like IETF for Internet protocols or W3C for Web standards,
VSO provides the rules — not the business.
Join the movement to establish cryptographic trust infrastructure for AI systems
"Aircraft have flight recorders. AI needs one too."
— VeritasChain Standards Organization
VAP Framework Specification is licensed under CC BY 4.0 International