Zemi North, Evidence Intelligence

AI Assurance

Test the claim, not the appearance of capability.

Independent examination of AI systems and the controls around them: what the evidence supports, what cannot be reconstructed, and who remains accountable.

Examination
Independent
Method
Evidence-led
Conclusion
Human-owned
AI ASSURANCE EXAMINATION Evidence converges on the claim.
01 / CLAIM Claim + intended use Assertion and intended use
DEFINED SYSTEM + DECISION BOUNDARY
02 / RECORD System evidence Data · prompts · logs
03 / OPERATION Behaviour + controls Tested in context
04 / AUTHORITY Human authority Roles · interventions · rights
EXAMINATION POINT Claim tested in context
05 / FINDING Bounded finding Human-owned · limitations stated
Record preserved Context tested Conclusion bounded

Assurance dimensions

Examination, not certification.

Zemi North provides evidence-based examination and assurance findings about defined systems, decisions, incidents, and claims. It does not issue accredited certifications or conformity certificates.

Claim What is asserted?
Evidence What can be inspected?
Control What constrained the system?
Accountability Who owns the outcome?
  1. Governance / Readiness

    AI governance and readiness

    Policies, risk assessments, procurement reviews, and governance controls designed to be operated, and defended, by real organizations.

    Typical question
    Are the controls proportionate to the system, its use, and the decisions it informs?
    Records
    Policies, risk assessments, system documentation, procurement records, roles, approvals, and monitoring evidence.
    Work product
    A scoped readiness or control assessment that identifies support, gaps, limitations, and recommended action.
    Discuss governance readiness
  2. Incident / Reconstruction

    AI incident investigation

    Reconstruct what a system did using available prompts, logs, outputs, configuration, and independent sources. Examine improper disclosure, misuse, or AI-assisted fraud.

    Typical question
    What did the system do, what authorized the action, and what remains unresolved?
    Records
    Prompts, outputs, tool calls, identity and access records, configuration, system logs, and surrounding evidence.
    Work product
    An incident timeline and findings report that separates system activity, human action, authorization, and uncertainty.
    Discuss an AI incident Forensic investigation practice ↗
  3. Decision / Traceability

    AI-assisted decision examination

    Determine whether an automated or AI-assisted decision can be traced, reproduced, explained, and supported by evidence.

    Typical question
    Can the decision be reconstructed and does the available record support the explanation given?
    Records
    Inputs, outputs, model and configuration records, decision rules, human interventions, approvals, and audit logs.
    Work product
    A decision reconstruction that identifies the evidence path, material gaps, and the limits of reproducibility.
    Discuss a decision examination
  4. Media / Authenticity

    Synthetic-media assessment

    Structured evaluation of authenticity claims and limitations. No single detector is treated as conclusive evidence.

    Typical question
    What does the available record support about the origin, alteration, or authenticity of the media?
    Records
    Original files where available, metadata, provenance records, platform context, comparison material, and corroborating sources.
    Work product
    An assessment of supported indicators, conflicting evidence, unresolved questions, and conclusion limits.
    Discuss a media assessment

Independent AI examination / Non-confidential enquiry

Before your organization relies, test what it can defend.

Begin with the system, decision, incident, or claim that requires independent examination.

Discuss AI assurance