Australian Government gets AI-Ready: new AI Technical Standard and launch of Gov AI

AI Technical Standard

The Technical Standard sets out a set of consistent requirements and best practices for the design, development, deployment, monitoring, and decommissioning of AI systems. It is designed to ensure the safe, ethical, and responsible use of AI, strengthen public trust, and align with evolving regulatory and technological landscapes.

The Technical Standard is designed to complete and operationalise existing policies and frameworks governing AI use in the public sector, including the:

The Technical Standard also emphasises the 'reuse' of agency policies, frameworks and practices, rather than introducing new processes or duplication.

The practices outlined in the Technical Standard take the form of:

Statements – describing ‘what’ needs to be done.

Criteria – each statement has at least one criterion to satisfy the statement, with each being criterion being marked either as:

required - driven by legislation, regulation and policies and ethics principles, and which agencies must satisfy to meet the Technical Standard. For example, agencies are ‘required’ to ensure that AI systems are auditable throughout their lifecycle (Criterion 9), or
recommended – which agencies should nevertheless seek to implement. For example, it is ‘recommended’ that agencies start with small AI model architectures and add complexity gradually (Criterion 75). This approach is considered best practice as it is intended to simplify debugging and reduce error, however it is not required.

Explanatory notes – which are provided for each criterion and are intended to offer guidance rather than serve as a comprehensive checklist.

The below table sets out, at a high-level, key details of the Technical Standard:

The Technical Standard combines established good practice for technology management with specific measures to address the unique risks, behaviours, and societal impacts of AI.

For example, the general principles of governance, accountability and transparency apply to all use of technology. However, there are additional requirements and practices that are specific to AI systems, such as AI bias, explainability, model management, and ethical considerations.

What do agencies need to do?

Review and map current AI practices: assess existing agency policies and processes against the Technical Standard to identify gaps and areas for improvement. The Technical Standard recognises that AI system development is an iterative process which requires ongoing review, assessment, testing and improvement. To ensure effective post-deployment improvements, it will be important for agencies to provide workforces with appropriate training on AI systems to enable them to provide timely feedback and identify issues after deployment.
Designate accountable officials: ensure clear assignment of responsibility for AI governance and compliance within the agency. This means assigning specialist roles to build and maintain AI capabilities, as well as the mandatory obligation for agencies to appoint an accountable official to be responsible for implementing the Policy for responsible use of AI in government.
Update data management frameworks: strengthen data quality, consent, and supply chain management processes in line with the requirements of the Technical Standard. Check contracts with suppliers to make sure this can be done through the supply chain.
Strengthen testing and monitoring: adopt robust testing, validation, and monitoring practices throughout the AI lifecycle, including post-deployment. Where you have a contacted supplier, check contracts to make sure this is part of the software maintenance obligations.
Engage with suppliers: where procuring third-party AI solutions from suppliers, ensure contractual arrangements reflect the requirements of the Technical Standard, in particular as they relate to transparency, audit and risk management.
Prepare for transparency and audit: consider whether any changes need to be made to your public AI transparency statement (remember these need to be updated regularly) and ensure comprehensive documentation and audit trails are maintained.
Plan for decommissioning: establish decommissioning plans for AI systems, including secure data destruction and stakeholder communication protocols. These plans need to clearly identify the system components being shut down and the reason for the decommissioning. Agencies must conduct an impact analysis and inform all affected parties, including providing support and information about any alterative systems.

Conclusion

The Australian Government is taking significant steps to expand and uplift the AI capability of the Australian Public Service, including through the launch of Gov AI and the new AI Technical Standard.

The Technical Standard is designed to provide agencies with a framework to navigate the unique risks associated with AI, enabling them to realise the benefits of AI technologies while maintaining public trust, ensuring compliance, and upholding ethical standards.

Interested in hearing more on Australia’s approach to AI?

Join us today at our annual Digital Future Summit, where Danielle Wood (Productivity Commission Chair) will discuss the Commission’s recent Interim Reports and recommendations including on harnessing data & digital technology and the government’s approach to embracing AI. Click here to register. It’s not too late!

Whole-of-lifecycle approach	The Technical Standard applies across the entire AI system lifecycle, including: discovery - design, data, training, evaluation operate - integration, deployment, monitoring retire - decommissioning. The Technical Standard is applicable regardless of whether an agency develops an AI system in-house or contracts an external provider to build or supply it.
Governance, accountability, and transparency	Agencies must designate specific roles for AI implementation and maintain clear governance structures. These roles could include data scientists and analysts, AI integration engineers, AI test engineers, ethics and compliance officers and domain exports. This is in addition to the accountable official for AI implementation, which is a position required by the existing Policy for responsible use of AI in government (see also Standard for accountable officials). End-to-end auditability is required, including documentation, traceability, and audit logging throughout the AI lifecycle (see also the Digital Transformation Agency’s AI model clauses, which require suppliers to ensure automatic logging of AI systems)
Risk management and ethics – including managing bias risks	Ongoing monitoring and evaluation of AI systems is required to identify and address unintended impacts. Agencies must systematically identify, assess, and manage bias throughout the AI system lifecycle to ensure compliance with federal anti-discrimination law. Agencies must establish a bias management plan, conduct thorough evaluations for systemic, human, statistical, and computational bias, and implement ongoing monitoring and mitigation strategies. Regular bias awareness training, disaggregated performance metrics, and mechanisms for user feedback and escalation are also required. Agencies must also ensure compliance with copyright law, information management legislation, data privacy and protection practices, and ethics practices.
Data management and quality	Requirements for data quality, validation, and provenance are set out, including data profiling, labelling, and bias mitigation. This is essential to ensure that AI systems are reliable, ethical, transparent and effective in delivering intended outcomes. Agencies must establish data supply chain management processes, considering consent, storage, access, retention, and destruction policies. Data sharing and integration should comply with relevant legislation and ethical frameworks. This includes processes for ‘data fusion’, which is a method to combine data from multiple sources to help AI systems produce more accurate output. Legislation and frameworks that agencies need comply with include: Data Availability and Transparency Act 2022, Guidelines on data matching in Australian Government administration and Data Ethics Framework.
Model development and testing	Agencies must define clear success criteria and select appropriate metrics (such as metrics relating to value-proposition, performance, bias, safety, reliability, adoption). Model training, validation, and selection must be rigorous, with requirements for explainability, bias evaluation, and continuous improvement. Testing strategies must address the probabilistic and non-deterministic nature of AI, meaning that an AI model does not guarantee a specific outcome and can produce different outputs even when provided with the same inputs. Strategies to address this include robust regression, scenario, and adversarial testing.
Deployment, monitoring, and decommissioning	Secure, incremental, and non-disruptive deployment strategies are required, with phased roll-outs, readiness verification, and change management protocols. Continuous monitoring of AI systems post-deployment is mandatory, covering performance, safety, reliability, human-machine collaboration, unintended consequences, transparency and explainability, costs, security and compliance. Structured decommissioning plans must be in place, including impact analysis, stakeholder communication, secure shutdown, and detailed documentation.

Australian Government gets AI-Ready: new AI Technical Standard and launch of Gov AI

Last week, the Australian Government released a comprehensive Technical Standard for the use of Artificial Intelligence (AI) across government agencies.

AI Technical Standard

What do agencies need to do?

Conclusion

Governance Solutions

Crisis Management

Innovation at Mallesons

Owl Advisory by Mallesons

Early Careers

Qualified Lawyers

Shared Services and Support

Brisbane

Canberra

Melbourne

Perth

Sydney