AI Execution Certification: The Certified Execution Record

The problem with traditional AI logs

Logs help engineers debug. They do not help organizations prove what happened. When a regulator, auditor, or customer asks "what did the AI actually do?", a log file is not a defensible answer.

• Mutable. Logs can be edited, truncated, or deleted after the fact. Nothing cryptographically binds the log to the execution it describes.
• Incomplete. Standard logs rarely capture the full picture: the exact prompt, model version, temperature, system instructions, and tool calls that produced a given output.
• Unverifiable. No third party can independently confirm that a log entry reflects what actually ran. You have to trust the system that wrote it.
• Not structured for audit. Regulatory and audit workflows benefit from structured, tamper-evident evidence of AI decision-making. Unstructured log files generally do not meet that bar.

For a detailed comparison, see Logs vs Certified Execution Records.

How Certified Execution Records solve this

A Certified Execution Record (CER) is a structured, tamper-evident bundle that captures the complete execution context of an AI operation and seals it with a certificateHash: a SHA-256 hash over all protected fields.

CERs are the fundamental unit of the NexArt AI Execution Integrity surface. They replace opaque logs with reproducible, auditable execution evidence.

What a CER captures

• Inputs: The exact prompt, system instructions, and any context provided to the model.
• Parameters: Model identifier, version, temperature, and configuration used during execution.
• Outputs: The complete model response, including structured content and metadata.
• Tool calls: Any external tool invocations made during execution, including function calls, API lookups, and retrieval results.
• Context signals: Execution environment evidence including active permissions, memory state, session scope, and any constraints the system operated under.
• Timestamps and versioning: When execution occurred and under which protocol version.

What a CER looks like

A CER is a portable JSON artifact. Here is a simplified example showing the key protected fields:

{
  "bundleType": "cer.ai.execution.v1",
  "protocolVersion": "1.2.0",
  "executionId": "exec_8f2a…d41c",
  "certificateHash": "sha256:4f31edabd565…",
  "timestamp": "2025-12-18T14:32:07.241Z",
  "input": {
    "prompt": "Classify this claim as approved or denied.",
    "systemPrompt": "You are a claims reviewer…",
    "model": "gpt-4o",
    "temperature": 0.1
  },
  "output": {
    "content": "Classification: APPROVED. Rationale: …",
    "finishReason": "stop",
    "tokensUsed": { "prompt": 842, "completion": 156 }
  },
  "toolCalls": [
    {
      "name": "lookupPolicyLimits",
      "arguments": { "policyId": "POL-29481" },
      "result": { "coverageMax": 50000 }
    }
  ],
  "attestation": {
    "attestationId": "att_c9e1…",
    "nodeSignature": "ed25519:…",
    "nodePublicKey": "…"
  }
}

The certificateHash is computed over all protected fields using deterministic canonical JSON serialization. Any modification to any field invalidates the hash. The attestation block is added when the CER is submitted to an independent attestation node.

Logs vs CERs at a glance

For the full comparison, see Logs vs Certified Execution Records.

Integrity guarantees

The certificateHash binds all protected fields into a single verifiable value. If any field is modified after sealing, the hash changes and verification fails.

Optionally, a CER can be submitted to an attestation node which verifies internal consistency and issues a signed receipt: an Ed25519-signed attestation that enables offline verification without contacting the node.

Verification is independent. Anyone can recompute the hash, check the signature, and confirm integrity without an API key, an account, or any dependency on NexArt. Use verify.nexart.io or the SDK.

When verifiable execution matters

CERs are most valuable where AI decisions carry consequences and may need to be reviewed, defended, or audited after the fact.

• Compliance and audit. Demonstrating to auditors exactly what an AI system did, with cryptographic proof the record is unaltered. See standards alignment.
• Third-party accountability. Providing verifiable evidence to customers, partners, or regulators about AI-driven outcomes.
• Regulated industries. Financial services, healthcare, insurance, and legal contexts where AI decisions carry material consequences. See regulated workflows.

Who uses CERs

Frequently asked questions

Getting started

NexArt provides the infrastructure to create, certify, and verify CERs through a simple API.

• Read the AI Execution Integrity specification for the full technical details.
• Follow the quickstart guide to create your first CER.
• Use verify.nexart.io to independently verify any CER.
• Review pricing. Execution is free. Certification is what you pay for.