Named accountability
A person or team is responsible for each significant AI use case across its lifecycle.
Human control and accountability
AI should support humans, not remove responsibility.
AI systems can support analysis, drafting, recommendations, pattern recognition, workflow automation and decision-making. The organisation and its responsible people remain accountable for how AI is used and for the outcomes it produces.
What human control means
People must be able to review, intervene and be accountable.
Review before use
Humans check AI outputs before high-impact, sensitive or customer-facing use.
Intervention
People can pause, override or stop an AI system that behaves unexpectedly or causes risk.
Transparency
Affected people receive appropriate information when AI use matters to them.
Testing and monitoring
AI systems are tested before deployment and monitored for bias, drift, misuse and harm.
Training
Workers understand AI limitations and know how to avoid over-reliance.
AI agents
AI agents need stronger control.
Agentic AI systems can combine generative AI with tools, data and workflows to complete multi-step tasks. These systems may operate with more autonomy than ordinary AI assistants, making permissions, access controls, logging, audit trails, testing and human approval more important.
Six principles
Practical principles for responsible AI control.
- Decide accountability
Know who is responsible across the AI lifecycle.
- Understand impacts
Assess effects on people, the organisation and the wider community before deployment.
- Manage risks
Use existing risk frameworks and dedicated AI controls.
- Share essential information
Tell users, workers and stakeholders what they need to know.
- Test and monitor
Review performance before and after deployment.
- Keep intervention possible
Maintain meaningful human control and the ability to stop unsafe use.