Release document 2 _ Redacted

Written by Claude Haiku 5.5 from the extracted text on 20 Oct 2026. It can miss or misread redacted and scanned pages. Check the source before relying on it.

The release is an email chain between Services Australia and the DTA's AI Governance team, dated 17 to 19 December 2025. Services Australia notified the DTA under the Policy for the responsible use of AI in government of a potentially high-risk AI use case: the AGDRP Additional Fraud Controls Model. This is a non-generative, supervised machine learning model, run on-premises without third-party software, that scores Australian Government Disaster Recovery Payment claims for fraud risk. Flagged claims go to trained analysts, with a human in the loop, and the model does not determine eligibility or fraud. The agency's Customer Committee approved the use case, and it sought advice from the Independent Advisory Board for Government Services. Services Australia will retest the risk rating with the DTA and may run a controlled soft-launch at the first 2026 AGDRP activation. The release flags reputational risk and staff concerns about job losses. The DTA acknowledged receipt and asked to be told when the use case is deployed.