AI execution evidence
Your AI made a decision.
Can you prove it?
When an auditor, regulator, customer, or court asks you to prove that your AI made this decision on these inputs, teams usually reconstruct it from logs, databases, and scattered systems. NexArt creates portable, independently verifiable evidence of what your AI actually executed, so you can answer before the question is asked.
Built for engineers, compliance teams, and regulated-industry operators.
Cryptographically sealed. Externally timestamped when certified. Publicly verifiable.
One SDK call. No infrastructure replacement. Sensitive data can remain private.
Why evidence is needed
When an AI decision is challenged, internal logs are not enough.
Logs can be edited, deleted or reconstructed. Inputs, parameters and outputs are often stored separately. That may support debugging, but it does not give an external reviewer independent evidence of what happened.
Customer dispute
A customer challenges an AI-generated decision or action.
Audit request
An auditor asks for evidence tied to the original execution.
Incident investigation
Your team must establish exactly what the system received and returned.
Regulated workflow
A material AI decision requires defensible execution-level records.
Certified Execution Records
Every recorded execution produces its own evidence package.
A lightweight SDK records the selected inputs, parameters, outputs and execution context at runtime. It seals them into a tamper-evident artifact called a Certified Execution Record.
The NexArt attestation node checks and signs the record. Any third party with access to the record can independently verify its integrity using the record, the signature and the node’s public key.
How it works
Execute. Certify. Verify.
Three distinct layers with distinct trust boundaries. Each can be adopted independently.
Execute
The SDK creates the record inside your environment.
Local creation · No certification fee · No infrastructure change
Certify
The NexArt attestation node checks and signs the record.
Attestation signature · Optional external timestamp · Defensible evidence
Verify
A third party recomputes the record hash and checks the signature.
No account · No API key · No dependency on NexArt-side data
Confidential execution
Keep the data private. Still prove the execution.
NexArt can replace raw inputs and outputs in the record with cryptographic commitments. Sensitive prompts, customer data and model outputs remain within the organisation while the integrity of the execution can still be independently verified.
Explore Confidential Execution- Raw inputs and outputs do not need to be published.
- Altered data will no longer match the commitment sealed into the record.
- Third parties can verify the integrity of the record without seeing the confidential content. When data is later disclosed, they can confirm that it matches the original commitment.
Where it holds up
Built for moments when “trust us” is not enough.
AI agent actions
Preserve verifiable evidence of the inputs, recorded tool calls and resulting actions across an agent workflow.
See AI agent evidenceCustomer-facing decisions
Respond to challenges involving lending, insurance, eligibility, pricing or recommendations.
Explore decision disputesApproval and policy workflows
Demonstrate which inputs, parameters and policy context were associated with a decision.
See policy workflow evidenceAI incident review
Preserve execution evidence before systems, models or logs change.
Review incident use casesAudit and assurance
Give reviewers a portable artifact they can inspect independently.
Explore audit evidenceMulti-step workflows
Group related execution records into one Project Bundle covering an agent session or business process.
Understand Project BundlesEvidence, not observability
Observability helps you inspect your system. NexArt helps somebody else verify it.
Logs and observability
- Designed for debugging and operations
- Controlled by the system operator
- Depend on databases and retention policies
- Describe what the system reports happened
NexArt execution evidence
- Designed for proof and independent review
- Tamper-evident and cryptographically signed
- Portable outside the originating system
- Lets another party verify record integrity
NexArt complements observability rather than replacing it. Keep your traces and logs for operations. Add NexArt where you need portable proof.
What verification proves
A verifier holding the record and the relevant public key can check its integrity even if the original operator or NexArt is unavailable.
NexArt proves
- The record has not changed since sealing.
- Recorded inputs, parameters and outputs remain cryptographically bound.
- The stated attestation node signed the record.
- Disclosed data matches the cryptographic commitments created for it.
- Execution records remain bound to their stated Project Bundle.
The record proves the integrity of what was captured. The reliability of capture depends on how the integration and execution environment are controlled.
NexArt does not prove
- The output is correct, safe or unbiased.
- The original source data was truthful.
- The selected policy was appropriate.
- Human oversight occurred unless it was recorded.
- The wider system is compliant or properly governed.
Hosted verification
Use verify.nexart.io for simple public verification. No account or login is required. This hosted path requires the verification service to be available.
Independent verification
Use the CLI, SDK or standard cryptographic libraries with the record and relevant public key. No NexArt account or live NexArt service is required.
npx @nexart/cli verify record.cer.jsonimport { verify } from '@nexart/ai-execution'
verify(bundle)Framework alignment
Supports evidence and record-keeping activities relevant to frameworks including the EU AI Act, ISO/IEC 42001, SOC 2 and NIST AI RMF. NexArt does not itself make a system compliant.
Get started
Create evidence before you need to defend the decision.
Start with one AI call. Produce a verifiable execution record, certify it through an attestation node and give reviewers a verification path that does not depend on your internal logs.
Free to create. Paid to certify. Public to verify.
Stay close
New articles, protocol updates, and product notes. No spam, no fluff.