Purpose before deployment
Use AI where it supports a defined human, service or organisational outcome. Do not treat adoption as an objective by itself.
Accountability
Name accountable owners for the use case, data, controls, operation and decisions affected by the system. Responsibility remains with people and organisations.
Proportionate risk management
Assess potential harm, affected groups, uncertainty and consequence. Apply stronger evaluation, approval and human oversight where impact is higher.
Privacy, security and data stewardship
Use data lawfully and minimise exposure. Protect models, prompts, interfaces and outputs. Make provenance, permission and retention part of solution design.
Transparency and human agency
Help users understand when AI materially shapes an interaction or decision, what its limits are and how to question, correct or escalate an outcome.
Evaluation and continuous improvement
Test before release, monitor in context, document material changes and provide routes to pause or withdraw systems that do not meet the intended standard.