-
The bill would cover Commonwealth automated actions that have, or could reasonably be expected to have, a significant legal or practical effect on a person, community or the environment. It would regulate automation already authorised under another Commonwealth law; it would not create a new power to automate decisions. Its rules would override inconsistent Commonwealth laws to the extent of the inconsistency.
-
Government bodies could not automate covered actions where the result is likely to be unlawful, requires human judgement, unduly influences a person's discretion, cannot be explained, unjustifiably harms human rights, or is likely to significantly harm Australia's democratic system or public confidence. Regulations could create exceptions for specified decisions requiring human judgement.
-
Before using covered automation, the responsible department would have to assess its risk and publish that assessment on a public register. Every assessment would need written approval from the department's accountable authority. Automation must be classed as high risk if the action could reasonably be expected to have a significant adverse effect on a person, community or the environment. High-risk assessments would also need Commonwealth Ombudsman approval; if the Ombudsman did not decide within two months, approval would be automatic.
-
Departments would have to review published risk assessments every 12 months for high-risk automationAutomation classified as high risk in a published risk assessment. It must receive that classification when the action could reasonably be expected to significantly harm a person, the community or the environment. and every three years for other covered automation. A revised assessment would go through the approval process again. High-risk assessments not updated within 18 months, and other assessments not updated within three years and six months, would be removed from the register.
-
Each automated action would need a named human role responsible for it. Where another Commonwealth law already requires someone to be notified of the action, that notice would also have to say automation was used, explain why the outcome occurred and how the computer was involved, identify the responsible officerThe human legally treated as having taken an automated action., and explain review rights.
-
High-risk automationAutomation classified as high risk in a published risk assessment. It must receive that classification when the action could reasonably be expected to significantly harm a person, the community or the environment. would need written legal advice and testing for accuracy, different outcomes across groups, and missing or out-of-range information. Department leaders would have to be satisfied that the program considers relevant information, ignores irrelevant information, does not improperly limit human judgement, treats people fairly and follows the law in force at the time. They would have to certify this in writing and record their reasons. Departments would also have to consult people or representative organisations likely to be disproportionately affected, as well as people or representative organisations who will interact with or administer the computer program.
-
If a private supplier operates or maintains high-risk automationAutomation classified as high risk in a published risk assessment. It must receive that classification when the action could reasonably be expected to significantly harm a person, the community or the environment., the Commonwealth would need an enforceable right to obtain the information required to oversee it; commercial confidentiality could not block access. A human would need power to replace an automated action when it was not correct or preferable. A separate complaints process would have to acknowledge complaints within five business days and provide a substantial response within 30 business days.
-
High-risk automated decisions could only be used where an affected person could request a fresh internal reviewA fresh examination of a decision within the responsible government body. by someone who did not use the original computer program. The reviewer could uphold or replace the decision. Reviews would be due within four weeks, or within two weeks for categories where a longer delay may cause serious and irreversible harm. Review by the Administrative Review TribunalThe independent tribunal that reviews certain government decisions. and a court would also have to remain available, although tribunal review could wait until internal reviewA fresh examination of a decision within the responsible government body. was finished.
-
The Commonwealth Ombudsman would maintain a free public register with information about high-risk systems, including how they work, their testing and whether a human can override them. The register could not contain personal information about an individual. Information could be withheld where publication would undermine the purpose or administration of the law behind the action, but the Ombudsman could require a written explanation. Departments would report each year on the number, kinds and proportion of high-risk actions replaced by humans. The Ombudsman would review a risk-weighted sample at least every 12 months and every high-risk system at least once every three years. Records would generally be kept for at least seven years.
-
The core bans would apply from the day after Royal Assent. Most other obligations would apply to automated actions occurring after a 12-month transition period. The operation of the legislation would also face an independent review within three years, and every three years after that.