In short
Risk appetite sets the line inside which teams move fast
- Risk appetite tells the organisation where the line between acceptable and unacceptable risk sits. Inside that line, teams can move as fast as they are able.
- The 5 categories of AI risk show boards what they agree to. They are autonomy, misuse for destruction, misuse for power, economic displacement and unknown risks.
- The Board AI Risk Register turns risk appetite into concrete thresholds: green (deploy), amber (review) and red (board approval).
- AI is now firmly on the Australian board and regulatory agenda. A board that defines its appetite early sets its own limits before an incident forces a decision.
Detail
Green, amber and red zones set who approves each AI use
Boards face the same tension with AI: move fast enough to compete and carefully enough to stay compliant. Some resolve it by restricting all AI use, which leaves them less competitive. Others approve everything and create exposure.
The Board AI Risk Register resolves this with explicit zones. The green zone covers internal productivity tools with no customer data and no automated decisions, which teams can deploy with their own approval. The amber zone covers customer-facing uses with human oversight and non-sensitive data, which need executive review and documented controls. The red zone covers automated decisions about customers, sensitive data and uses with regulatory implications. These need board notification and independent review.
The zones rest on 5 categories of AI risk. They are adapted from those Anthropic chief executive Dario Amodei sets out in his January 2026 essay 'The Adolescence of Technology'. Each category becomes a question about the proposed system. Autonomy risk asks how much decision-making authority the system has. Misuse for destruction asks whether it could cause harm if it malfunctions or is compromised. Misuse for power asks whether it concentrates decision-making inappropriately. Economic displacement asks whether it changes the workforce in ways the organisation must manage. Unknown risk asks what is not yet known about the system's behaviour at scale.
Each proposed deployment is rated against the 5 categories, and the combined rating sets its zone and the governance that applies. A legal issue or an unacceptable failure mode puts a use in the red zone, whatever its combined rating. The method takes much of the subjective debate out of risk appetite.
Explicit risk appetite can also speed up deployment. When teams know what sits in the green zone, they stop waiting for permission on low-risk uses. They can deploy more AI because they know what they may approve themselves.
The Australian regulatory context adds weight to all of this. The Australian Securities and Investments Commission (ASIC) published Report 798 in October 2024. It states that directors' and officers' duties of reasonable care and diligence 'extend to the adoption, deployment and use of AI'. The Privacy and Other Legislation Amendment Act 2024 added new privacy policy disclosure duties for automated decisions. A board that defines its own risk appetite now decides how the organisation will operate under these rules.
Commercial impact
Risk appetite errors are costly in both directions
Getting risk appetite wrong is costly in both directions.
Consider a board that is too cautious. It rejects every AI proposal because it needs more information. Meanwhile, competitors deploy AI in customer service and pricing, and market share moves to them. That revenue may not come back.
Now consider a lender that is too aggressive. It automates lending decisions without adequate bias testing, and the system disadvantages applicants from certain postcodes. The lender then faces discrimination complaints, remediation costs and damage to its reputation. A risk assessment before deployment would have called for bias testing.
Without a risk appetite framework, a board cannot see the upside or the downside of a proposal clearly.
The 5 categories make the assessment complete. Many boards think about AI risk mainly as data privacy and cyber security. They can miss autonomy risk, where AI makes decisions without human review. They can also miss economic displacement, which affects entry-level roles, and unknown risk, which covers how a model behaves in unusual cases.
From 10 December 2026, Australian Privacy Principle (APP) 1.7 and APP 1.8 set privacy policy disclosure rules for some automated decisions. Our article on what APP 1.7 requires sets out the test in full. A board that has already defined its risk appetite will find those rules easier to work with.
In practice
Illustrative cases of placing AI uses in risk zones
Illustrative composite: a financial services board sets 3 zones with written criteria. Before, every proposal needed executive sign-off whatever its risk, so teams deployed very few tools. With the zones written down, green-zone tools no longer wait for a board cycle.
Illustrative composite: a board places an AI credit assessment tool in the red zone. It rates high on autonomy risk, because it makes automated decisions, and on economic displacement, because it replaces analyst work. Before full deployment, the board requires bias testing, documentation of how the tool reaches its decisions and a supervised trial alongside the current process. It accepts the time this adds to the schedule.
Illustrative composite: a healthcare provider finds that its AI triage system prioritises patient callbacks without clinician review. Nobody had assessed that autonomy risk. Placing the system in the red zone means clinician review is added before it continues.
Illustrative composite: a mid-market retailer sets its appetite as green for internal operations AI, amber for customer-facing recommendations and red for automated pricing. Its operations team can then move quickly on warehouse automation. Pricing decisions have competition law implications, so they still receive full review.
What to do
Steps to define and apply AI risk appetite
- Set up the Board AI Risk Register with green, amber and red zones, and rate every AI proposal against the 5 risk categories.
- Write down which uses teams may deploy without executive approval, and let them proceed.
- Before any red-zone AI is deployed, require bias testing and documentation of how it reaches its decisions.
- Review risk appetite on a set cycle, such as quarterly, as regulation, AI capability and competition change.
- Define your risk appetite before 10 December 2026, when APP 1.7 and APP 1.8 apply.
Last reviewed
Put this topic against your own numbers, owners and timelines.