To: Australia should establish a Federal framework making enduring human agency a protected public-interest principle for consequential AI and software systems.
Ensuring Human Agency with AI
KEEP AI WORKING FOR US
A proposed GetUp campaign for Human Agency by Design
AI is already delivering real benefits to Australians. It helps people learn, create, communicate, solve problems and make decisions. Today, most people experience AI as a helping agent: we ask, AI assists, and we remain in control.
That relationship is worth protecting. The emerging threat is not AI itself. It is the possibility that increasingly capable AI will progressively acquire authority over decisions and actions that affect people, while human control becomes weaker without a clear public decision ever having been made.
Why is this important?
THE THREAT: HUMAN CONTROL CAN BE LOST ONE DELEGATION AT A TIME
AI development is moving beyond assistance toward delegated AI agency and increasing autonomy. A person may first ask AI for an answer. Later, that person may authorise an AI agent to carry out a task. Increasingly autonomous systems can then plan, sequence actions, use other software and act with less contemporaneous human direction.
HELPING AGENT → DELEGATED AGENCY → AUTONOMY
The danger does not require a dramatic 'AI takeover'. Human agency can be eroded incrementally: one decision is delegated, then another, then a process, then an organisational function. The human may remain formally responsible while practical authority has migrated to the system.
This is an emerging and increasingly important public-interest risk. The fact that today's AI is useful and generally user-directed must not obscure the consequences of tomorrow's more autonomous systems.
WHAT IS HUMAN AGENCY BY DESIGN?
Human Agency by Design means treating human authority as an intentional, protected property of every consequential AI-enabled system — not as an optional ethical feature added after the technology has been built.
It has two inseparable dimensions:
• Governance intention — society and government establish who may delegate authority to AI, what limits apply, who remains accountable, how people can challenge decisions, and when human intervention is mandatory.
• Architectural intention — those requirements are built into identities, permissions, credentials, data, APIs, workflows, agent frameworks, audit trails, intervention mechanisms and the underlying software infrastructure.
The principle is simple: increasing machine capability must not automatically become increasing machine authority.
WHY ACT NOW?
Once organisations and governments become dependent upon autonomous systems, restoring human control may be technically, economically and institutionally difficult. The opportunity to establish the boundary is greatest while humans still determine how AI is incorporated into the digital infrastructure of society.
History provides a warning. Robodebt and the Post Office Horizon scandal demonstrate the human and institutional cost that can arise when automated system outputs acquire authority while affected people lack effective means to understand, challenge or correct decisions. AI could magnify such risks through greater scale, speed, complexity and delegation.
The objective is therefore preventive: protect human agency before it becomes difficult to recover.