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 138

Macquarie Technology Group

Submitter group: Companies.

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.

Macquarie Technology Group, a data-centre and sovereign-cloud provider, argues sovereign AI requires ownership, control, jurisdiction and assured access beyond physical location. Proposes risk-tiered safeguards for high-consequence AI, procurement and capability measures, and targeted reforms for demonstrated legal gaps. Welcomes foreign investment alongside coordinated regulation and domestic capability development.

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. It argues sovereign AI is defined too narrowly when equated with hosting data or compute in Australia, since location alone does not guarantee control.

    Current understanding among Australian policy makers risks defining AI sovereignty too narrowly. Housing data, compute or inference capability in Australia is important, but physical location alone does not guarantee sovereign control.

  2. Argues Australia should distinguish essential sovereign control, desirable Australian participation and sufficient trusted international access, rather than choosing complete independence or dependency.

    Australia should reject the false choice between complete technological independence and complete dependency. It should adopt a middle path that identifies where sovereign control is essential, where meaningful Australian participation and leverage are desirable, and where trusted international access is sufficient.

  3. It argues a foreign AI company establishing an Australian base is valuable but not itself sovereign Australian capability.

    Attracting a US hyperscaler or AI company to establish an Australian base would be valuable but should not itself be characterised as sovereign Australian AI capability.

  4. It argues supply-chain security for AI should extend beyond hardware to model weights, software, data and serving infrastructure.

    Supply-chain security must therefore extend beyond hardware to model weights, software components, training and evaluation data, orchestration tools, identity systems and model-serving infrastructure.

  5. It argues existing frameworks already apply to aspects of AI, so demonstrated gaps should be identified and frameworks adapted before creating general regulation.

    Existing privacy, consumer, cyber-security, critical-infrastructure, administrative-law and sectoral frameworks already apply to aspects of AI use. The immediate task should be to identify demonstrated gaps, clarify accountability and adapt existing frameworks before creating duplicative general regulation.

  6. It says Commonwealth procurement requirements could set sovereignty, hosting, auditability and continuity standards for government AI.

    First, Commonwealth procurement requirements could specify sovereignty, hosting, auditability and continuity standards for AI systems used by government agencies.

Recommendations

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

  1. Adopt a whole-of-stack definition of sovereign AI assessing ownership, control, jurisdiction, assured access, data, infrastructure, software, IP, skills and governance.

    Adopt a whole-of-stack definition of sovereign AI that assesses ownership, control, jurisdiction, assured access, data, infrastructure, software, intellectual property, skills and governance.

  2. Map nationally important AI use cases and determine whether each needs sovereign control, Australian participation or trusted foreign access.

    Map nationally important AI use cases and determine where each requires sovereign control, Australian participation or trusted foreign access.

  3. Establish a risk-tiered sovereign AI governance framework for high-consequence government, national-security and critical-infrastructure applications.

    Establish a risk-tiered sovereign AI governance framework for high-consequence government, national-security and critical-infrastructure applications.

  4. Allow risk-justified Australian hosting of weights/copies, inference and serving infrastructure; Australian credential, key and emergency-access control; regulatory/audit access, local records and incident/remediation capability, change notification and continuity protections.

    Depending upon the risk and function, relevant requirements could include: • Australian-hosted model weights or secured copies of relevant weights; • Australian-hosted inference and model-serving infrastructure; • Australian control of access credentials, operational keys and emergency access arrangements; • enforceable Australian regulatory and audit access; • local retention of safety, incident and model-operation records; • Australian incident-response and remediation capability; • advance notification of material model, service or governance changes; • continuity, portability, recovery and service-restoration obligations;

  5. For defined critical functions, consider restricting remote suspension/withdrawal and testing continued operation during disruption or legal conflict.

    restrictions on remote suspension or withdrawal for defined critical functions; and • tested arrangements allowing continued operation during disruption or legal conflict.

  6. Expand hosting certification or establish equivalent AI assurance covering infrastructure, operations, data, models, weights and governance, calibrated to use-case sensitivity and consequences.

    Second, the Hosting Certification Framework could be expanded, or an equivalent AI assurance framework established, to classify AI environments by the degree of infrastructure, operational, data, model, weight, and governance sovereignty they provide. Requirements could be calibrated to the sensitivity and consequence of the use case.

  7. Establish an Australian AI Capability and Economic Benefit Framework requiring tenderer disclosures of stack ownership/control, R&D/employment locations, IP/licensing, Australian data treatment, local supply chains, commercialisation/exports and foreign-jurisdiction continuity risks.

    Macquarie recommends an Australian AI Capability and Economic Benefit Framework. Tenderers should disclose ownership and control of critical stack components; location of research, development and skilled employment; ownership and licensing of intellectual property; treatment of Australian data; Australian supply-chain participation; commercialisation and export pathways; and continuity risks arising from foreign jurisdiction.

  8. Require explanations against transparent capability, risk and economic-benefit criteria when agencies choose foreign AI over a competitive Australian capability.

    The Commonwealth should implement an “if not, why not?” requirement. Where a competitive Australian capability exists, an agency selecting a foreign alternative should explain the decision against transparent capability, risk and economic- benefit criteria.

  9. Use procurement, co-investment and challenge-based programs to build Australian capability in higher-value layers and support commercialisation and exports.

    Use procurement, co-investment and challenge-based programs to build Australian capability in higher-value layers and support commercialisation and exports.

  10. Require Commonwealth-funded/procured high-consequence AI to identify Australian governance responsibility, auditable model/data controls, enforceable incident/remediation arrangements, log/evidence access and credible continuity arrangements.

    Macquarie recommends that Commonwealth-funded or procured high- consequence AI demonstrate identified Australian responsibility for governance, auditable model and data controls, enforceable incident and remediation arrangements, access to relevant logs and evidence, and credible service- continuity arrangements.

  11. Require effective Australian jurisdictional control and redress for high-consequence government AI.

    Require high-consequence government AI to demonstrate effective Australian jurisdictional control, governance, auditability, incident response, redress and continuity.

  12. Treat Australian data as a strategic asset and support domain-specific models in sectors of Australian expertise.

    Treat Australian data as a strategic asset and support domain-specific models in sectors of Australian expertise.

  13. Establish clearer policy on how different classes of data are stored, processed and protected, linked to the assurance level required of the underlying AI and cloud environment.

    Government should establish clearer policy for how different classes of data are stored, processed and protected, and link those rules to the assurance level required of the underlying AI and cloud environment.

  14. Apply the Commonwealth definition of an Australian business consistently when assessing domestic capability and economic benefit.

    Apply the Commonwealth definition of an Australian business consistently when assessing domestic capability and economic benefit.

  15. Welcome frontier AI and data-centre investment with coordinated, proportionate, technology-neutral and internationally competitive regulation.

    Welcome frontier AI and data-centre investment while ensuring regulation is coordinated, proportionate, technology-neutral and internationally competitive.

  16. Align Commonwealth/State planning, energy, water, environment and infrastructure policy; avoid charges making viable projects uneconomic, recognise flexible energy solutions and provide timely nationally significant infrastructure pathways, preserving safety/environmental standards.

    Australia should not respond by weakening safety or environmental standards. It should respond through coordinated, proportionate and predictable regulation. Commonwealth and State governments should align planning, energy, water, environmental and infrastructure settings; avoid charges that make otherwise viable projects uneconomic; recognise flexible energy solutions; and provide timely pathways for nationally significant digital infrastructure.

  17. Lead sustained public communication explaining AI/data-centre benefits for productivity, services, cyber security, infrastructure, jobs and regions, alongside safety, security, energy, water and planning safeguards.

    Government must therefore lead a sustained public conversation about the benefits and obligations arising from AI and data-centre growth. It should explain how the sector contributes to productivity, public services, cyber security, critical infrastructure, skilled employment, and regional opportunity, while also explaining safeguards for safety, security, energy, water, and planning.

  18. Use greater Australian sovereignty and jurisdictional control to strengthen public confidence and social licence.

    Use greater Australian sovereignty and jurisdictional control to strengthen public confidence and social licence.

  19. Determine which AI capabilities are important enough to national resilience that continuity of access must be assured, and which dependencies can be managed through diversification, contractual safeguards or trusted access.

    Government should determine which AI capabilities are sufficiently important to national resilience that continuity of access must be assured, and which dependencies can be managed through diversification, contractual safeguards or trusted access.

  20. Assess critical component ownership/control, modification or withdrawal powers, jurisdiction, regulator evidence access and continuity if foreign law or policy changes.

    Government should therefore assess who owns and controls model weights, credentials and serving systems; who can modify, update, restrict or withdraw the capability; which jurisdiction governs the intellectual property and operator; whether Australian regulators can obtain evidence; and whether continuity can be assured if foreign law or policy changes.

  21. Avoid basing Australian policy on assumed continuing US government–AI industry alignment on safety, regulation, deployment or expansion.

    Australia should avoid assuming that the US Government and US AI companies will necessarily remain aligned in their approaches to safety, regulation, deployment or expansion.

  22. Consider Commonwealth procurement standards for sovereignty, hosting, auditability and continuity, with tenderer disclosure of weights, inference, credentials, IP, dependencies and incident-record ownership/control.

    First, Commonwealth procurement requirements could specify sovereignty, hosting, auditability and continuity standards for AI systems used by government agencies. Tenderers could be required to disclose ownership and control of model weights, inference environments, credentials, intellectual property, service dependencies and incident records.

  23. Consider sector-specific obligations for AI supporting critical infrastructure, nationally significant systems or essential government functions, using existing critical-infrastructure and cyber-security frameworks to identify wider disruption consequences.

    Third, sector-specific obligations could apply where AI supports critical infrastructure, nationally significant systems or essential government functions. Existing critical-infrastructure and cyber-security frameworks provide a basis for government to identify systems whose disruption would have wider consequences.

  24. Consider Australian control, knowledge-transfer, safety, continuity and commercialisation conditions in government funding/co-investment agreements, particularly for models, inference, cyber-security and national-security applications.

    Fourth, government funding or co-investment agreements could include Australian control, knowledge-transfer, safety, continuity, and commercialisation conditions. This would be particularly relevant where public funds support models, inference capability, cyber-security platforms or other national-security applications.

  25. Consider high-consequence AI legislation requiring Australian audit records, significant-incident reporting, regulator evidence access, human oversight/remediation and essential-function continuity when offshore access is restricted.

    Fifth, legislation for high-consequence AI could require operators to maintain Australian audit records, report significant incidents, provide regulators with access to evidence, demonstrate human oversight and remediation arrangements, and ensure that essential functions can continue if offshore access is restricted.

  26. Keep governance mechanisms risk-based, avoiding blanket Australian-owned hosting of foreign providers' weights; reserve stronger controls for justified safety, security, infrastructure resilience or government continuity needs.

    These mechanisms should be risk based. They should not require all foreign AI providers to store all model weights in Australian-owned data centres. They should permit stronger controls where public safety, national security, critical-infrastructure resilience or continuity of government justify them.

  27. Extend AI supply-chain security beyond hardware to weights, software, training/evaluation data, orchestration, identity and model-serving infrastructure.

    Supply-chain security must therefore extend beyond hardware to model weights, software components, training and evaluation data, orchestration tools, identity systems and model-serving infrastructure.

  28. Identify demonstrated legal gaps, clarify accountability and adapt existing frameworks before duplicative general regulation.

    The immediate task should be to identify demonstrated gaps, clarify accountability and adapt existing frameworks before creating duplicative general regulation.

  29. Consider targeted reform where existing law inadequately covers high-consequence AI transparency, auditability, incident reporting, autonomous-outcome responsibility, continuity or regulator access, directed to defined risks and functions.

    Targeted legislative reform may nevertheless be justified for high-consequence AI where existing law does not adequately provide transparency, auditability, incident reporting, responsibility for autonomous outcomes, continuity or regulator access. Reform should be directed to defined risks and functions rather than AI generally.

  30. Assess public AI investment's strategic and economic returns, including assured access/control, resilience, Australian IP, R&D, skills, jobs, supply chains, commercialisation, exports and Australian-owned firm growth.

    Applied to AI, public investment should seek two returns: • Strategic return: assured access, Australian governance, cyber security, resilience, continuity and control over nationally important functions. • Economic return: Australian intellectual property, research and development, skills, employment, supply chains, commercialisation, exports and growth of Australian- owned firms.

  31. Evaluate procurement's whole value and contract for measurable outcomes rather than choosing suppliers solely by nationality.

    This is not a proposal to select suppliers on nationality alone. It is a proposal to evaluate the whole value created from procurement and contract for measurable outcomes.

  32. Seek international AI investment across data centres, deployment/inference, regional operations, research partnerships, advanced computing and ecosystem development.

    Australia should actively welcome investment from US and other international AI companies. Frontier firms can bring capital, skills, infrastructure, research partnerships and access to advanced models. Australia should seek to become a preferred platform for their expansion across the Asia Pacific. 16.1 That ambition should include data centres, model deployment and inference, regional service operations, research partnerships, advanced computing infrastructure and ecosystem development.

  33. Prioritise support for domain models, sovereign inference/deployment, AI cyber/national security, assurance/governance tools, cloud/orchestration and software aligned with Australian legal, regulatory and operational needs.

    Government support should help Australian businesses accumulate capability in layers that create durable intellectual property, technical know-how, market power and export potential. 13.1 Priority opportunities include domain-specific models trained on Australian data; sovereign inference and deployment environments; AI-enabled cyber-security and national-security applications; trusted assurance and governance tools; cloud and orchestration capability; and software platforms aligned to Australian legal, regulatory and operational requirements.

  34. Ensure Australian authorities have visibility and enforceable authority over high-consequence AI systems operating in Australia.

    If the chief executive of a leading frontier developer considers future models capable of presenting risks serious enough to justify independent monitoring and a slower pace of development, Australian policy should ensure Australian authorities have visibility and enforceable authority over high-consequence systems operating here.

Read the original

Open in the interactive guide