SOAI-PDCA Continuous Improvement Framework
The industry-standard methodology for ongoing AI safety governance, combining Safety Oversight AI with the proven Plan-Do-Check-Act cycle for continuous improvement and regulatory compliance.
What is SOAI-PDCA?
SOAI-PDCA combines AI-powered safety oversight with the Deming Cycle (PDCA) to create a continuous improvement loop for AI governance that meets global regulatory requirements including EU AI Act, NIST AI RMF, and TC260.
Safety Oversight AI (SOAI)
The proposed Council has a designed 33-seat roster and a target threshold of 23/33, strictly greater than two thirds. It is not live. The latest experiment measured rho=1 and n_eff=1, so independence and fault tolerance are not demonstrated; the evidence is published on the Refutation Ledger.
- Vendor independence is a measurement target, not a current claim
- Proposed 23/33 review threshold
- Human analyst oversight
PDCA Cycle (Deming Cycle)
The proven four-phase methodology for continuous improvement, adapted specifically for AI safety governance and regulatory compliance.
- Iterative improvement process
- Evidence-based decision making
- Continuous compliance monitoring
The Four-Phase SOAI-PDCA Cycle
PLAN
Define objectives, identify risks, establish compliance requirements
DO
Implement controls, deploy safeguards, execute mitigation strategies
CHECK
Monitor performance, assess compliance, collect incident reports
ACT
Analyze results, implement improvements, update policies
Detailed Phase Documentation
Step-by-step implementation guides for each phase of the SOAI-PDCA cycle, with downloadable templates and real-world examples.
Phase 1: PLAN
Establish objectives and processes necessary for AI safety
Key Activities:
- Identify AI system scope and boundaries
- Define safety objectives and success criteria
- Map applicable regulatory frameworks (EU AI Act, NIST, TC260)
- Conduct initial risk assessment
- Establish baseline compliance requirements
- Define roles and responsibilities
- Create implementation timeline
Required Documentation:
SOAI Integration:
A future 33-seat Council could assist in the PLAN phase by:
- Analyzing AI system architecture for potential risks
- Recommending applicable compliance frameworks
- Identifying similar systems and historical incidents
- Suggesting best practices from industry standards
Getting Started with SOAI-PDCA
Follow these steps to implement the framework in your organization
