Auditable AI Execution
How Certified Execution Records and signed attestation receipts map to evidence requirements for ISO/IEC 42001, SOC 2, and NIST.
If your organization is preparing for or maintaining SOC 2, ISO 42001, or NIST-aligned compliance, deploying AI in regulated workflows raises a specific question: how do you produce evidence of what your AI systems actually did, in a form your auditors will accept?
NexArt is built to produce that evidence. Certified Execution Records and signed receipts are tamper-evident, independently verifiable audit artifacts that map directly to the control evidence requirements of SOC 2, ISO 42001, NIST AI RMF, and NIST SP 800-53.
Important
NexArt is an evidence layer: it produces tamper-evident audit artifacts (Certified Execution Records + signed receipts) that can be verified independently.
NexArt provides verifiable execution evidence. It does not by itself make an organization compliant.
Mappings below are guidance; auditors may require additional controls and evidence beyond what NexArt provides.
Designed to align with AIEF v0.2.x control objectives (Levels 1–4) for verifiable AI execution records. NexArt is one implementation; AIEF is vendor-neutral and open for public comment.
Supports Level 4-style outcomes: multi-step chain integrity, tool/dependency evidence, and independent verification.
As AI systems move into regulated and high-stakes workflows, integrity of execution records becomes a prerequisite for defensibility, not just observability.
What NexArt provides
- Certified Execution Record (CER), a single JSON record that binds inputs, parameters, and outputs together.
- Certificate hash, a tamper-evident seal over the record contents.
- Node stamp (signed receipt), optional Ed25519 signature over a receipt. Verify offline using the node's published public keys. Receipt includes: receipt, signature, attestorKeyId.
- Deterministic replay (deterministic workloads only), for deterministic workloads (e.g., Code Mode), rerunning the snapshot reproduces the same output.
Note: For probabilistic models, NexArt verifies the recorded artifact; it does not attempt to reproduce tokens.
- Verification reason codes, PASS / FAIL plus machine-readable reason codes explaining what didn't match.
- Redacted exports + provenance, you can remove sensitive fields and still produce a verifiable artifact, while preserving the original historic hash as reference.
Legacy note: older records may show a legacy attestation (integrity-verifiable but not offline signature-verifiable).
How NexArt fits in your compliance program
NexArt is infrastructure, not a SaaS dependency. The records it produces are verifiable independently of NexArt itself: an auditor reviewing them does not need to trust NexArt, only the cryptographic primitives and the published public keys.
NexArt does not currently hold SOC 2 or ISO certification. Independent verifiability covers the record itself; operator-level questions (how signing keys are held, rotated, and governed) are addressed through documented controls and, where a customer requires it, formal certification. It does not replace operator assurance.
Scope boundary. NexArt covers execution and inference. It does not cover training-time, model development, or retirement and decommissioning. It supports the record-keeping and traceability slice of ISO 42001, NIST AI RMF, and EU AI Act Article 12. It does not perform risk classification, conformity assessment, or the overall governance program, and it does not replace risk assessments, model approvals, model inventory, change management, or monitoring. It sits underneath those activities as one input an auditor consumes.
Control outcomes
What auditors typically want to see, and how NexArt helps.
| Outcome | NexArt capability | Evidence artifact |
|---|---|---|
| Tamper-evident decision trail | Certificate hash over stable, deterministic record representation | CER JSON + verify PASS/FAIL |
| Proof that logs weren't edited post-hoc | Hash mismatch detection + reason codes | Verification report + reason codes |
| Third-party integrity stamp | Signed receipt (Ed25519) + public key discovery | Signed receipt (receipt + signature + attestorKeyId) + node keys document |
| Reproducible execution (where deterministic) | Deterministic snapshots + replay | Code Mode CER + replay output |
| Traceability of automated decisions | Workflow/run IDs + step chaining (agent workflows) | RunBuilder chain + final hash |
| Privacy-preserving audit sharing | Redacted export + provenance | Redacted CER + meta.provenance |
| Long-term verifiability | Frozen hashing semantics + backward-compatible verification | Old CERs still verify with new SDKs |
Typical audit workflow
- Record, issue a CER for each AI execution (inputs, parameters, outputs bound together with a certificate hash).
- Stamp, optionally request a signed receipt from a NexArt node to add an independent, offline-verifiable integrity seal. The stamp confirms integrity checks and signs a receipt; it does not validate decision correctness.
- Archive, export the CER (redacted if needed) as a portable JSON artifact for long-term retention. CERs are portable JSON artifacts suitable for long-term retention and independent verification.
- Verify, at audit time, re-verify the certificate hash and stamp independently using public keys or the browser verifier.
What you can claim when verification PASSes
- Protected fields in the record have not been modified since issuance under the declared scheme.
- If a signed receipt is present, the signature verifies against the node's published public key.
What you cannot claim
- That the decision was correct, fair, or unbiased.
- That the model would output the same tokens if re-run today.
SOC 2 mapping
| SOC 2 area (TSC) | Evidence for your audit | Example evidence artifact |
|---|---|---|
| Security | Supports integrity of audit evidence by making records tamper-evident | CER JSON + verification result |
| Security | Adds independently verifiable stamps on recorded evidence | Signed receipt + node keys doc |
| Processing Integrity | Shows exactly what was recorded as inputs/params/outputs for automated decisions | CER snapshot fields |
| Processing Integrity | Detects post-hoc modifications with clear reason codes | FAIL + reason code + details |
| Confidentiality | Enables redaction while preserving verifiability of the exported artifact | Redacted CER + redacted hash + provenance |
| Availability (indirect: audit portability) | Provides portable audit artifacts that can be verified offline | Exported CER + local verify |
| Privacy (supporting) | Helps limit disclosure through selective redaction and proof separation | Redaction + provenance guidance |
What this means for your audit: when your SOC 2 review examines your AI controls, you can present verifiable execution records that the auditor validates independently using NexArt's published public keys, without needing to trust NexArt or your internal logs.
SOC 2 requires broader controls (access, monitoring, incident response). NexArt helps with integrity evidence.
NIST mapping
Family-level mapping (SP 800-53 families).
| NIST area | Evidence for your audit | Example evidence artifact |
|---|---|---|
| AU · Audit & Accountability | Produces audit records that are tamper-evident and independently verifiable | CER JSON + verification PASS/FAIL |
| AU · Audit & Accountability | Supports independent verification without trusting the originating app | Offline verification + node keys |
| SI · System Integrity | Detects unauthorized changes to recorded executions | Reason codes + mismatch details |
| CM · Configuration Management | Version-pinned semantics (protocol + SDK) help reproduce the meaning of historic records | Protocol version + sdkVersion |
| SC (supporting: audit confidentiality via redaction) | Separates proof from sensitive content (receipt/signature vs raw inputs) | Signed receipt + redacted exports |
| IR · Incident Response (supporting) | Provides clear forensic artifacts for what changed and when (or what verified) | Verification details + provenance |
NIST AI RMF
| RMF function | Evidence for your audit | Example evidence artifact |
|---|---|---|
| GOVERN | Supports accountability via immutable decision records and stamps | CER + signed receipt |
| MAP | Captures decision context (inputs/params/model) for traceability | Snapshot fields |
| MEASURE (supporting) | Enables auditing and evaluation of recorded outcomes | CER exports for review |
| MANAGE (supporting) | Helps demonstrate controls around change detection and incident forensics | Reason codes + provenance |
What this means for your audit: NexArt produces the kind of audit records and forensic artifacts NIST control families anticipate, with cryptographic verifiability that supports independent third-party review.
ISO/IEC 42001 mapping
Theme-based mapping to AI management system requirements.
| ISO/IEC 42001 theme | Evidence for your audit | Example evidence artifact |
|---|---|---|
| Traceability of AI outputs | Binds input/parameters/output into a single immutable record | AI CER bundle |
| Monitoring & measurement | Produces verification outcomes + reason codes for integrity checks | verify() result + codes |
| Accountability & governance | Enables independent verification using third-party stamps and public keys | Signed receipt + keys doc |
| Change / incident handling | Helps investigate integrity disputes with precise mismatch explanations | FAIL + details + provenance |
| Documentation & retention | Portable JSON artifacts that can be archived and verified years later | Exported CER + frozen semantics |
| Supplier/provider management (supporting) | Captures provider/model identifiers and parameters at the time of decision | provider/model/params fields |
What this means for your audit: an ISO 42001 audit examining traceability, monitoring, and accountability of AI outputs can consume NexArt's Certified Execution Records directly as evidence, archived in portable JSON suitable for long-term retention.
ISO 42001 requires an AI management system (policies, roles, risk treatment). NexArt provides evidence artifacts that support audits.
EU AI Act mapping
The EU AI Act is primarily a governance and risk-management regime. Most of its requirements concern how an organisation classifies, oversees, validates, and documents its AI systems. NexArt does not address those governance requirements. Where NexArt helps is narrower and specific: it produces tamper-evident, independently verifiable AI execution evidence that supports the Act's record-keeping and traceability obligations (Article 12), and the broader expectation that an organisation can later demonstrate, to an external party, what its AI did.
| EU AI Act area | Evidence NexArt supports | Example evidence artifact |
|---|---|---|
| Record-keeping (Article 12) | Provides tamper-evident records of AI executions that can be retained and independently verified over the system lifecycle | CER JSON + verification PASS/FAIL |
| Traceability of high-risk AI operation | Binds input, parameters, model, and output into a single immutable record for a given execution | AI CER bundle + certificate hash |
| Human oversight evidence (Article 14, supporting) | Helps demonstrate what was executed and what a human reviewer was presented with, without claiming the oversight itself was adequate | CER snapshot fields + signed receipt |
| Post-market monitoring (supporting) | Supplies portable, verifiable execution artifacts that can be examined if an incident or dispute arises later | Exported CER + provenance |
| Provider / deployer accountability (supporting) | Captures provider and model identifiers and parameters at the moment of execution | provider / model / params fields |
| Demonstrability to regulators and auditors | Produces evidence that a third party can verify independently, without trusting the operator's own logs | Offline verification + node public keys |
What this means for your audit: if a regulator or auditor examining EU AI Act readiness asks not just whether governance processes existed, but what a specific AI execution actually did and whether that record was altered afterward, NexArt provides evidence that stands on its own. It does not replace the risk management, human oversight, or conformity assessment the Act requires.
The EU AI Act requires governance, risk classification, human oversight, and conformity assessment that NexArt does not provide. NexArt supplies execution-integrity evidence that supports specific record-keeping and traceability obligations only.
What NexArt does not cover
NexArt is an evidence and integrity layer, not a full compliance program. It does not replace:
- IAM / access controls and least privilege
- Encryption / key management and secure storage
- Model risk management (MRM) policies and approvals
- Bias / fairness evaluation and monitoring
- Human oversight and operational controls
- Vendor due diligence and contractual compliance
- Decision correctness or domain-specific compliance approval
- EU AI Act risk classification, conformity assessment, and governance obligations
Try it
- Issue an AI record (nexartaiauditor.xyz)
Generate a sample Certified Execution Record without any setup. See exactly what record format your audit evidence will take.
- Verify independently in browser (verify.nexart.io)
Paste a certificate hash from any CER, including ones you generated yourself, and verify it cryptographically. No account, no login, no NexArt dependency.
- Browser-only verifier on NexArt.io(/verify)
Same verification surface integrated into the main site for in-context evaluation.
- Public node keys (well-known endpoint)
Published Ed25519 public keys used to verify signed receipts. Verification can be performed entirely offline against these keys.
Related
- Certified Execution Records, What CERs are and how they work
- AI execution integrity, Full execution surface specification
- AI execution verification, Verify CERs independently
- Logs vs Certified Execution Records, Why traditional logging falls short
- Deterministic execution records, Hash and canonicalization guarantees
From the blog