Purpose and necessity
Use AI where it produces a meaningful benefit and define the decision or workflow it supports.
Human authority
Responsible AI in Practice
Responsible AI is not a disclaimer attached after deployment. It is expressed through choices about purpose, information, models, authority, evaluation, transparency and what intelligent systems are permitted to do.
Principles
Responsibility is expressed through design decisions
AFA’s existing responsible-AI principles remain the foundation, now extended to the full intelligence environment.
Human accountability
Assign clear ownership for decisions, approvals, exceptions and escalation.
Information protection
Match data, knowledge and model access to the sensitivity and purpose of the workload.
Proportional risk
Apply stronger controls as the consequences, autonomy and information sensitivity increase.
Transparency and maturity
Describe systems and product maturity accurately; do not imply certainty or availability that does not exist.
Evaluation and monitoring
Test behaviour before use and monitor material changes in models, sources, prompts and workflows.
Inclusion and accessibility
Design experiences that can be understood and used by the people they are intended to serve.
Lifecycle responsibility
Govern the system through design, deployment, operation, change and retirement.
Authority
Define what the person is actually responsible for
Human oversight is not meaningful when a person is nominally “in the loop” but lacks the time, information or authority to intervene.
For agents and Physical AI, Operational Sovereignty adds a direct question: what is the system permitted to do without approval, and what must remain under human authority?
Operating framework
Turn principles into an operating system for responsibility
Document responsibilities that can survive beyond a pilot.
Purpose
Define the outcome and why AI is appropriate.
Boundaries
Define information, tools, models and actions that are allowed.
Evidence
Define evaluation, logging, provenance and documentation requirements.
Accountability
Define approval, escalation, change control and who can stop the system.
Continue the journey