AI Inquiry Atlas

An unofficial, independent guide to the public records of the Joint Select Committee on Artificial Intelligence. Not connected with the Parliament of Australia or the committee. Its makers also made a submission (no. 481; see About). Summaries and scores are written by AI and can be wrong: check the original before relying on them.

Submission 10

Australian National Audit Office

Submitter group: Government.

An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.

Open in the interactive guide

Summary

Read the original submission for the full text.

Describes ANAO audits finding the ATO's AI arrangements partly effective and IP Australia's AI use in patent rights largely effective, with monitoring weaknesses. Recommends improvements to both agencies' governance, ethics, assurance, evaluation and information management, and risk-scaled safeguards across government. The cover letter describes continuing IT control weaknesses as significant business risks.

Key points

Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.

  1. Says the ATO's arrangements to support its adoption of AI were partly effective.

    The ATO has partly effective arrangements in place to support its adoption of AI.

  2. Says 74 per cent of ATO AI models in production lacked completed data ethics assessments.

    For its AI models, the ATO has not complied with the requirements of this framework (74 per cent of AI models in production did not have completed data ethics assessments).

  3. States that there is no one-size-fits-all approach to AI governance and that governance and assurance should match the level of risk.

    There is no one-size-fits-all governance approach to support the adoption of AI. At the entity level, governance and assurance arrangements should be commensurate with the level of risk.

  4. Says IP Australia's strategic framework did not consistently link AI initiatives to defined business outcomes, while inventory, committee roles and use-case ownership remained in progress.

    The strategic framework does not consistently link AI initiatives to defined business outcomes, and the AI inventory, role clarity across committees and use-case ownership remain in progress.

  5. Says IP Australia's monitoring and reporting of AI impacts was partly effective, with practices varying across tools and scheduled reviews not fully implemented.

    IP Australia's monitoring and reporting of the impact of its use of AI is partly effective. Monitoring practices varied across patent rights tools, and scheduled review processes under the new governance framework are not yet fully implemented.

  6. The ANAO reports that IT control findings continue to be its most significant area of audit findings, and that weaknesses can expose entities to unauthorised access or data leakage.

    information technology (IT) control findings continue to be the most significant area of audit findings3. Weaknesses in controls in this area can expose entities to an increased risk of unauthorised access to systems and data, or data leakage.

Recommendations

Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.

  1. ATO: align automation and AI strategy implementation with enterprise-wide requirements.

    The ATO align implementation arrangements for the automation Paragraph 2.29 and AI strategy with enterprise-wide requirements.

  2. ATO: define and communicate enterprise-wide AI structures and governance, including model- and system-level accountabilities and responsibilities.

    The ATO clearly define and communicate enterprise-wide Paragraph 2.44 organisational structures and governance arrangements supporting its adoption of AI, including defining accountabilities and responsibilities at the model and system level.

  3. ATO: review misuse-of-data-and-analytics enterprise risk under its risk framework and appetite; update and incorporate AI-related controls.

    The ATO review the misuse of data and analytics enterprise risk in Paragraph 2.61 accordance with its enterprise risk management framework and risk appetite, and update and incorporate controls relating to the impact of AI on this risk.

  4. ATO: specify AI reproducibility and auditability requirements, integrate data ethics into other processes, complete production-model ethics assessments, and monitor, assure and report framework implementation.

    The ATO improve its arrangements in support of the design, Paragraph 2.92 development, deployment and use of AI that aligns with ethical principles by: (a) specifying requirements relating to AI reproducibility and auditability; (b) ensuring the data ethics framework is integrated into other ATO processes; (c) completing ethics assessments for the AI models in production; and (d) introducing monitoring, assurance and reporting arrangements over the implementation of its data ethics framework.

  5. ATO: develop and implement AI model design, development, deployment and assurance policies and procedures; align model practices with existing relevant policies and procedures.

    The ATO develop and implement policies and procedures to support Paragraph 3.30 the effective design, development, deployment and assurance of AI models. Where relevant ATO policies and procedures exist, the ATO ensure that the design, development and deployment of AI models aligns with these.

  6. ATO: establish performance measurement and evaluation for its automation and AI strategy.

    The ATO establish performance measurement and evaluation Paragraph 4.22 arrangements for its automation and AI strategy.

  7. ATO: manage information to support transparency and accountability in AI adoption.

    The ATO ensure that its approach to managing information supports Paragraph 4.31 transparency and accountability with respect to its adoption of AI.

  8. Government entities should scale AI governance to risk, scale and maturity, adapting it as AI use and complexity grow.

    There is no one-size-fits-all governance framework for AI. Governance should be commensurate with the risk, scale and maturity of an entity’s AI use. As the use and complexity of AI grows, an entity should adapt its governance arrangements.

  9. Define AI roles and responsibilities at enterprise and system levels.

    There should be clearly defined roles and responsibilities for AI — at both the enterprise level and the AI system level.

  10. Entities should establish arrangements to identify, manage and escalate emerging AI risks.

    Entities should have a focus on establishing arrangements for identifying, managing and escalating new and emerging risks.

  11. Support AI Ethics Principles with practical implementation arrangements, potentially including processes, guidance, monitoring and assurance.

    AI Ethics Principles should be supported by implementation arrangements that ensure principles are applied in practice. These arrangements could include processes, practical guidance, monitoring and assurance.

  12. Use risk-based policies, procedures and assurance for AI design, development and deployment.

    There should be risk-based policies, procedures and assurance arrangements to support the effective design, development and deployment of AI.

  13. Design monitoring and evaluation to measure outcomes and continually improve AI adoption at enterprise and system levels.

    Monitoring and evaluation arrangements should be designed to measure outcomes and to support an entity to continually improve the suitability, adequacy and effectiveness of its adoption of AI — at both the enterprise level and the AI system level.

  14. Report internally on AI adoption and use to senior management and the accountable authority.

    There should be internal reporting on the adoption and use of AI within the organisation to senior management and the accountable authority.

  15. Entities should adopt AI transparently and accountably, retaining records demonstrating this throughout design, development, deployment and continuous monitoring.

    Entities should ensure that they adopt AI in a transparent and accountable manner. Records should demonstrate transparency and accountability in the design, development, deployment and continuous monitoring of AI systems.

  16. IP Australia: review AI cyber security governance controls and ensure structured assessment, reassessment and authorisation of AI systems.

    IP Australia review its current controls for cyber security governance Paragraph 4.5 in relation to AI and ensure there is a structured approach to assess, reassess and authorise AI systems.

  17. IP Australia: establish risk-based AI implementation monitoring and reporting to support strategic oversight of benefits, costs and risks and inform planning and strategy.

    IP Australia establish clearly defined, risk-based monitoring and Paragraph 4.24 reporting arrangements on AI implementation, to enable strategic oversight of benefits, costs and risks that inform planning and future strategy.

  18. Regularly reassess and adapt AI governance frameworks and artefacts when scaling up AI use.

    When scaling up from experimentation to increased use of AI, governance frameworks and artefacts should be regularly reassessed and adapted to ensure they remain fit for purpose.

  19. Regularly review ICT controls to ensure they continue addressing risks.

    Controls over ICT systems provide assurance that digital tools are being used as intended. Reviews of these controls should occur regularly to ensure that controls continue to address risks.

Read the original

Open in the interactive guide