1-3.4 | Introduction, purpose, vision and objectives of the Framework | 0 |
3.5.1.1 | Transparency and Explainability: explain how AI applications reach decisions in terms people understand | 0 |
3.5.1.10 | Beneficial AI: promote the common good and well-being | 0 |
3.5.1.11 | Cooperation and Openness: foster multi-stakeholder cooperation and public understanding | 0 |
3.5.1.12 | Sustainability and Just Transition: mitigate societal and environmental impacts and leave no one behind | 0 |
3.5.1.2 | Reliability, Robustness and Security: operate correctly over time with the right data, handle errors and resist attack | 0 |
3.5.1.3 | Fairness: treat individuals in similar groups fairly, without favouritism, discrimination or harm | 0 |
3.5.1.4 | Diversity and Inclusion: respect the interests of all stakeholders and involve diverse users | 0 |
3.5.1.5 | Human Oversight: scale human intervention to the severity of ethical issues | 0 |
3.5.1.6 | Lawfulness and Compliance: act within all applicable laws, regulations, standards and guidelines | 0 |
3.5.1.7 | Data Privacy: honour the PD(P)O data protection principles and individuals' rights over their data | 0 |
3.5.1.8 | Safety: never compromise physical safety or mental integrity | 0 |
3.5.1.9 | Accountability: own the moral implications and name an accountable party | 0 |
3.5.2 | Operate the three-lines AI governance structure with defined roles from IT Board to data scientists | 0 |
3.5.3 | The six-stage AI Lifecycle and its alignment with the system development lifecycle | 0 |
4.1.1.1 | Set an AI strategy aligned with organisational goals and the Ethical AI Principles, with policies, metrics and trained staff | 0 |
4.1.1.2 | Keep up with AI standards and regulations and embed them in processes | 0 |
4.1.2.1 | Manage the AI portfolio: inventory, categorise and prioritise projects | 0 |
4.1.2.2 | Decide oversight needs at the outset, plan quality management and assign roles and resources | 0 |
4.1.3.1 | Maintain a technology roadmap for AI and data use and assess data needs | 0 |
4.1.3.2 | Procure AI services ethically: requirements, procurement plan, no lock-in, supply chain and IP, and third-party risk assessment | 0 |
4.1.4.1 | Define business objectives and weigh the benefits and risks of using AI | 0 |
4.1.4.2 | Design the solution: assess model suitability, set the degree of human intervention by risk, and plan explainability communications | 0 |
4.1.4.3 | Extract data with integrity: central repositories, validation, assurance of reliability and representativeness, annotation standards, documentation, transparent ETL and error handl | 0 |
4.1.4.4 | Pre-process data fairly and privately: security, privacy impact assessment, anonymisation or masking, removal of harmful content, secure work area, outliers, fairness testing, clas | 0 |
4.1.4.5 | Build models rigorously: validate assumptions, set risk-sensitive objectives, select inputs lawfully without leakage, control overfitting, use trusted training sources and defend a | 0 |
4.1.5.1 | Verify, validate and test thoroughly before deployment, including edge cases | 0 |
4.1.5.2 | Assume failure: layered mitigations, business rules, output checks and a disclosure document | 0 |
4.1.5.3 | Track mistakes and give users an easy feedback channel | 0 |
4.1.5.4 | Ensure repeatability, traceability and auditability | 0 |
4.1.6.1 | Monitor data and model performance: criteria, quality checks, anomaly detection and retraining | 0 |
4.1.6.2 | Provide operational support: service levels, user feedback and a support knowledge base | 0 |
4.1.6.3 | Continuously review and comply: independent review, management oversight, a termination process and IT Board consultation for high-risk applications | 0 |
5.1 | Complete the AI Application Impact Assessment: risk gating, assessment questions, practice references and impact considerations, with go or no-go decisions by PSC/PAT and IT Board | 0 |
5.2 | Review the assessment at the four lifecycle checkpoints, annually and on major change, and retain copies | 0 |
A-D | Appendices: glossary, references, impact assessment template and AI strategy template | 0 |