Accountability
Calibrated to AI
AIC certification is structured around how your organisation actually relates to AI in consequential decisions — not a one-size-fits-all compliance checkbox. Choose the Division that reflects your reality.
Your relationship with AI determines your certification path
The AIC Five-Division Framework recognises that organisations sit at very different points on the human-AI accountability spectrum. A hospital where clinicians sign off every AI diagnostic is fundamentally different from a platform where algorithms operate autonomously with periodic human review.
AIC certification is calibrated to your actual relationship with AI. Each Division has distinct requirements, monitoring obligations, and primary accountability KPIs — because generic compliance frameworks miss the nuance that matters.
We make decisions. Humans make them.
AI assists. Humans decide.
AI decides. Humans review patterns and cases.
AI operates. Systems and humans monitor outcomes.
We build AI. Others use it to make decisions.
The Five Divisions
Each Division maps to a distinct accountability model. Find yours and see exactly what AIC certifies.
Sovereign
“We make decisions. Humans make them.”
No AI in consequential decisions — verified annually
Organisations making no use of AI in consequential decisions.
Human accountability structures documented, no undisclosed AI in use, POPIA-compliant human data processing. Shadow AI audit confirms no automated decision systems.
Sovereign Assessment — point-in-time audit + annual renewal.
Prove you are fully human-accountable before regulators, clients, or media ask.
Supervised
“AI assists. Humans decide.”
Human override rate is the primary KPI.
AI generates recommendations; a named human makes every consequential decision.
Bank using AI credit model where loan officer decides. Hospital using AI diagnostic where clinician signs off. Employer using AI CV screening where recruiter approves shortlist.
Every AI system registered. Override process tested and evidenced. Decision records include both AI recommendation and human decision. Explanation mechanisms exist. Rejection communications meet dignity standards. Appeal processes functional. AI involvement disclosed to affected persons.
Full Pulse monitoring.
Reviewed
“AI decides. Humans review patterns and cases.”
Periodic human review rate is the primary KPI.
AI makes operational decisions; humans conduct periodic reviews and investigate flagged cases.
Lender with automated credit decisions + compliance officer reviewing weekly flags. HR tech platform auto-screening applications + recruiter reviewing rejected candidates weekly.
AI systems registered with documented risk categories. Periodic human review schedules documented and evidenced. Escalation protocols for flagged decisions functional. Bias testing conducted quarterly. Correction SLA ≤ 10 business days. Full transparency to users.
Pulse monitoring.
Monitored
“AI operates. Systems and humans monitor outcomes.”
Drift detection and aggregate outcome pattern monitoring.
AI operates autonomously with continuous technical monitoring; humans monitor aggregate metrics and investigate anomalies.
E-commerce AI recommendations monitored by algorithm team. AI fraud detection monitored by security team. AI route optimisation monitored by operations management.
All AI systems registered with documented purpose and risk category. Continuous technical monitoring in place (drift detection, performance metrics). Annual human review documented. Users informed they are interacting with AI. Anomaly escalation path exists.
Pulse monitoring focused on drift detection and aggregate pattern alerts.
Artificial
“We build AI. Others use it to make decisions.”
Product accountability architecture completeness.
Organisations that develop, train, and sell AI systems or models to other organisations. Their accountability is upstream — they are responsible for the accountability architecture their customers' decisions rest on.
SA LLM company selling to banks. Credit scoring SaaS selling to lenders. AI-powered medical diagnostic tool provider. HR tech company selling AI hiring tools.
AI product has documented accountability architecture for downstream users. Product includes human override capabilities. Explanation mechanisms built into the product. Bias testing conducted on training data and documented. Transparency disclosure published for each product. Process exists for receiving and responding to downstream incident reports.
Builder Certification — product-level certification. Each AI product sold receives its own certification. Analogous to CE marking or ISO product certification.
Division 5 certification does NOT replace the obligation of the Division 5 company's customers to hold their own AIC certification.
Your Certification Journey
Five steps from self-assessment to certified — with ongoing Pulse monitoring for organisations in Divisions 2, 3, and 4.
Self-Assessment
Complete AIC's diagnostic questionnaire to identify your Division classification based on how your organisation relates to AI in consequential decisions.
Gap Analysis
AIC conducts a structured gap analysis against the accountability requirements for your Division, identifying what evidence and controls are needed.
Evidence Review
Submit documentation of your human accountability structures, AI system registrations, override processes, and monitoring mechanisms.
Certification Audit
An AIC-accredited auditor conducts the formal assessment. For Sovereign and Builder certifications, this is a point-in-time audit.
Certification & Pulse
Upon passing, receive your AIC Division certificate. Supervised, Reviewed, and Monitored organisations enter continuous Pulse monitoring.
Ready to Certify Your Accountability?
Start with a diagnostic conversation. AIC will help you identify the right Division, understand the gap, and design a certification pathway that fits your organisation.
Enquire About Certification