Carrier-bound AI evidence
X-Road AI Evidence Lab
The lab studies whether governed data-exchange infrastructure can carry and witness independently verifiable evidence about AI-agent actions. X-Road is the first worked case.
The claim is deliberately narrow. X-Road does not certify AI behavior or prove that an output is true. It can bind an exchange to signatures, timestamps, routing context, and logs while a separate evidence package remains independently verifiable.
Two-layer evidence surface
Agent layer
Evidence package
Payload hash, verifier result, policy context, and package signature.
Binding
AEP hash in exchange
The carrier can move a reference to evidence without becoming the evidence verifier.
Carrier layer
X-Road exchange
Security Servers provide governed transport, signatures, timestamps, and message-log evidence.
Together, the two layers create an audit surface that neither layer provides alone: the carrier can witness that an exchange happened, while the evidence package can still be checked offline.
Reproducibility bundle
Public code, data, evidence packages, checksums, and lab records deposited on Zenodo.
Practitioner survey
A short anonymous survey for X-Road practitioners, open until 7 July 2026.
Current public lab
What has been demonstrated
The first public bundle uses X-Road development tooling and synthetic services to bind an Agent Evidence Package hash to a real X-Road exchange between two Security Servers. It is a reproducibility lab, not a production assurance report.
The bundle excludes production X-Road traffic, private deployment material, non-public operator records, production keys, and live operational data. It is designed so the lab can be inspected without crossing that boundary.
Next direction
X-Road 8 and dataspaces
X-Road 8 “Spaceship” is publicly framed as a move toward dataspaces, including the Dataspace Protocol, Decentralised Claims Protocol, Gaia-X Trust Framework compatibility, improved cloud compatibility, monitoring, and metrics. These features make the next lab question sharper: how should AI-agent evidence interact with participant credentials, policy negotiation, and carrier observability?
Public X-Road material: X-Road 8 vision, dataspaces approach, and development roadmap.
X-Road 8 beta smoke lab
Repeat the toy AI-service exchange on public X-Road 8 tooling and record what changes for deployment, logging, and evidence binding.
Credential bridge lab
Model service, verifier, and operator claims as verifiable credentials without collapsing agent evidence into carrier trust.
Dataspace policy-context lab
Bind an AI-agent exchange to data-sharing policy state, contract negotiation context, and verifier acceptance rules.
Monitoring and failure-mode lab
Measure what carrier monitoring can reveal, then replay stale, mismatched, and tampered evidence packages.
Boundaries
What this does not claim
- It does not claim that X-Road certifies AI behavior.
- It does not claim legal compliance with the EU AI Act or eIDAS.
- It does not use production X-Road traffic or private deployments.