Governance
AI in lending: what ASIC, NCCP and responsible lending actually mean for automation
By Saby Saxena1 min read
In short
Australian regulation doesn't ban AI in lending, it bans uncontrolled automation. Where decision support ends, and what a regulator-friendly AI looks like.
AI adoption in Australian lending comes with a question every executive eventually asks:
“Is this actually allowed?”
The short answer: yes, with guardrails.
What Australian regulation actually says
Under NCCP, ASIC guidance, and responsible lending obligations:
- Credit decisions must be defensible
- Processes must be explainable
- Consumers must be treated fairly
- Accountability must rest with licensed entities
Nowhere do regulations ban AI.
What they prohibit is uncontrolled automation.
Decision support vs decision making
This distinction matters.
AI is well-suited for:
- Data extraction
- Policy referencing
- Risk flagging
- Scenario analysis
- Customer communication
AI is not suited for:
- Final credit approvals
- Unsupervised hardship decisions
- Unexplained declines
Why early AI tools triggered concern
Many early AI lending tools:
- Used opaque models
- Couldn’t explain outputs
- Lacked audit trails
- Didn’t map to policy
This made compliance teams uncomfortable, and rightly so.
What a regulator-friendly AI looks like
A compliant AI system:
- Keeps humans in the loop
- Logs every action
- References policy explicitly
- Supports judgment, rather than replacing it
- Can be paused, audited, and overridden
Agent-based architectures excel here.
The path forward
Responsible AI in lending isn’t about avoiding innovation. It’s about embedding governance into the system itself.
See it on your own scenarios
onebank runs the work in between. Your team approves every decision.