AI CORE / DOCUMENTATION

Publish evidence, not system internals

Public output must have concrete value for an assessor while respecting security, privacy and the agreed scope.

Permitted public subset

Publish the requirement, method, scope, result, time of the performed observation, safe supporting evidence and limitations. Use a public snapshot identifier for the assessed state. Exact technical links may remain in the operator's protected storage.

What must not appear in public output

Secrets, credentials, personal data without a lawful basis, internal addresses and paths, server names, operational configurations, internal component versions, raw logs, complete prompts and private documents. An allowlisted schema helps constrain the scope, but type validation alone cannot guarantee that free text contains no sensitive information.

Redaction without distortion

Sanitized evidence should retain the substance of the observation and state what was removed or cannot be publicly examined. A sensitive finding may be described at a safe level, but it must not be relabeled as success. Protected details are provided only to an authorized assessor through an agreed channel.

No central collection

AI CORE provides static references. System evidence is not uploaded to this service, no login is used and no central audit log repository is created. The operator of the assessed system is responsible for access to its evidence.