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 363

Mr Michael Sanderson

Submitter group: Individuals.

Position labels are not shown for individual submitters.

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.

Argues that the Commonwealth should fund, own and operate independently operable sovereign AI and data centres as permanent public infrastructure, with private providers supplementing it. Emphasises real-resource constraints, public value, workforce transition, data control and accountable government AI. Supports consequence-based obligations and a regulatory grid, while leaving an AI Act open for cross-sector gaps.

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. Argues that the Commonwealth should permanently own and operate a sovereign AI and data centre capability for public purpose.

    The Commonwealth should establish and permanently maintain a sovereign AI and data centre capability that it owns and operates for public purpose.

  2. Argues that contractual access can be extremely useful without constituting sovereign capability.

    The critical distinction is between access and possession because contractual access can be extremely useful without constituting sovereign capability.

  3. Argues that AI policy should start from desired outcomes and distinguish private cost reduction from national productivity.

    AI policy should begin with the outcomes Australia wants rather than begin with adoption and assume that more AI necessarily produces more national productivity. A system can lower one company's labour costs while degrading service quality, transferring risk, increasing monopoly rents or consuming scarce resources whose alternative use would produce greater public value.

  4. Proposes national AI and data-centre resource planning covering electricity, water, land, skilled labour, training, materials, semiconductors, communications and strategic supply chains.

    A national AI and data centre strategy should consequently contain a real resource plan covering electricity, firming, transmission, water, land, skilled labour, training, construction materials, semiconductors, fibre and strategic supply chains.

  5. Argues that AI used for legally consequential Commonwealth functions must be understandable, auditable and explainable despite commercial confidentiality.

    Commercial confidentiality cannot override those obligations because a system the Commonwealth cannot adequately understand, audit or explain is unsuitable for a legally consequential public function.

  6. Argues that small businesses using embedded AI should not be required to act as auditors of systems they cannot inspect.

    Government should therefore resist frameworks that require every small business using AI enabled accounting, customer service or productivity software to become an amateur model auditor.

Recommendations

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

  1. Directly employ and retain AI researchers, software engineers, cyber specialists, data architects, systems engineers, electrical/mechanical specialists, model evaluators, privacy/procurement professionals and other staff needed to operate and adapt sovereign capability.

    The Commonwealth should directly employ and retain AI researchers, software engineers, cyber specialists, data architects, systems engineers, electrical and mechanical specialists, model evaluators, privacy specialists, procurement professionals and other staff required to understand, operate, test, maintain and adapt the system.

  2. Plan periodic hardware renewal, open/interoperable architecture where practicable, diverse supply, justified strategic inventories, secure maintenance and ongoing assessment of minimum capability for operational independence.

    Capability planning should therefore include periodic hardware renewal, open and interoperable architectures where practicable, diversified supply arrangements, strategic inventories where justified, secure maintenance capability and continuing assessment of the minimum national capability required to preserve operational independence.

  3. Make public compute available to universities, public research organisations and eligible smaller Australian enterprises where public purpose and capacity permit, on transparent, non-discriminatory public-interest or cost-recovery terms.

    The public system should also make appropriate capacity available to universities, public research organisations and eligible smaller Australian enterprises where doing so serves a defined public purpose and capacity permits. Access should be transparent and non-discriminatory on appropriate public interest or cost recovery terms so that the infrastructure expands innovation and competition without becoming an indiscriminate subsidy to private users.

  4. Protect public-capability governance against patronage, capture and stagnation through independent technical governance, privacy/security obligations, asset retention and periodic public reviews; protect genuinely classified functions through specific security arrangements.

    Governance of the public capability should itself be designed against political patronage, vendor capture and technological stagnation. It should operate under a clear statutory public interest mandate with independent technical governance, parliamentary reporting, strong privacy and security obligations, transparent procurement, asset retention requirements and periodic public capability reviews, while genuine classified functions can remain protected through specific security arrangements.

  5. Distinguish private cost reduction from national productivity; assess useful output, resources, distributional effects and what happens to labour and capacity released by automation.

    The present inquiry should therefore distinguish private cost reduction from national productivity and ask what useful output increases, what resources are consumed, who gains, who loses and what happens to labour and productive capacity released by automation.

  6. Differentiate policy support for adoption building Australian productive capability from adoption creating recurring payments and dependence on concentrated foreign providers.

    Policy support should therefore distinguish adoption that develops Australian productive capability from adoption that primarily creates another recurring payment and strategic dependency upon a concentrated foreign provider.

  7. Retain publicly financed strategic assets unless a clearly demonstrated public purpose justifies another arrangement.

    Publicly financed strategic assets should remain public assets unless a clearly demonstrated public purpose justifies another arrangement.

  8. Permit public contributions to private data-centre costs only for separate, transparent public benefits with corresponding public value.

    Public contribution should remain possible where separate and transparent public benefits justify it, but the public should receive corresponding value rather than simply subsidising a private input.

  9. Phase sovereign compute development with additional generation, firming, transmission and skills to expand capacity rather than compete for existing scarcity.

    The sovereign Commonwealth system should meet an even stronger standard because public ownership permits compute, energy and workforce planning to be integrated. Its development should be phased with additional generation, firming, transmission and skills so that sovereign AI expands the underlying productive base rather than merely becoming another large claimant upon existing scarcity.

  10. Assess water-intensive data centres in constrained regions at catchment level against agricultural, ecological, community and other competing uses.

    Water intensive data centres in constrained regions should therefore be assessed at catchment level against agricultural, ecological, community and other competing uses rather than treated as though cooling water has no alternative value.

  11. Assess private approvals and public projects against the national resource plan so cumulative commitments remain efficiently supplyable.

    Private approvals and public projects should be assessed against that national picture so that fragmented decisions do not cumulatively commit resources that Australia cannot efficiently supply.

  12. Require large data centres to identify flexible workloads before connection and, where technically/operationally feasible, contract for non-essential load reduction during defined system emergencies.

    Large data centres should therefore identify flexible workloads before connection and, where technically and operationally feasible, maintain contractual arrangements capable of reducing non-essential load during defined system emergencies. Such obligations should be established before connection when system requirements can be incorporated transparently rather than after a facility becomes a large entrenched load.

  13. Disclose major projects’ peak/average demand, network augmentation, added supply/firming, backup, water/cooling, heat/noise and demand reduction; protect genuinely sensitive details without concealing material effects on essential infrastructure.

    Assessment of major projects should disclose peak and average electricity demand, attributable network augmentation, additional supply and firming, backup arrangements, water use, cooling technology, heat and noise impacts and realistic demand reduction capability. Genuine security or commercially sensitive technical details can be protected, but commercial confidentiality should not conceal a project's material effect upon essential public infrastructure.

  14. Separate essential sovereign public workloads from flexible research/general compute, reducing lower-priority processing during serious system stress while protecting critical public, security and emergency functions.

    The Commonwealth sovereign capability should itself distinguish essential workloads from flexible research and general compute. Lower priority processing should be capable of reduction during serious system stress while critical public, security and emergency functions remain protected.

  15. Locate regional public compute nodes only where electricity, communications, security, water and workforce conditions justify them, potentially colocating research/training to create lasting capability.

    The public sovereign capability provides an opportunity for geographically redundant nodes where electricity, communications, security, water and workforce conditions justify them. Public research and training institutions can be developed alongside appropriate nodes so that resilience investment also leaves lasting regional capability rather than a stand-alone industrial load.

  16. Assess regional infrastructure allocation against technical suitability, resilience, national security, permanent jobs, training, local procurement, shared infrastructure and locally retained value.

    Regional allocation should nevertheless remain technically disciplined because placing infrastructure in an unsuitable location merely for geographic distribution would waste resources. The test should combine resilience and national security with permanent employment, training, local procurement, shared infrastructure and value retained in the surrounding region.

  17. Where governments assist private regional data centre development, use public benefit agreements addressing permanent jobs, apprenticeships, local procurement, TAFE and university relationships, shared infrastructure and access to useful compute where appropriate.

    Where governments assist private regional data centre development, public benefit agreements should address permanent jobs, apprenticeships, local procurement, TAFE and university relationships, shared infrastructure and access to useful compute where appropriate.

  18. Keep certification proportionate, independently governed and based on demonstrated reduction of material risk, so businesses that profit from certification cannot create mandatory standards that generate captive demand for their own services.

    Certification should likewise be proportionate, independently governed and based on demonstrated reduction of material risk. Businesses that profit from compliance, certification or assurance should not be allowed to create mandatory standards that primarily generate captive demand for their own services.

  19. Keep private access to sensitive Commonwealth information specific, lawful, minimised, logged, auditable and purpose-limited, with no implied external commercial-model training right.

    Sensitive Commonwealth information should therefore default to publicly controlled infrastructure, while private access should be specific, lawful, minimised, logged, auditable and limited to the purpose for which it is required. Procurement of a service should never create an implied entitlement to use Commonwealth data to train external commercial models.

  20. Retain sovereignty over derivative models, onward transfers, intellectual property, deletion, foreign legal exposure and continuing availability of public data.

    Sovereignty also requires control over derivative models, onward transfer, intellectual property, deletion, foreign legal exposure and continuing availability.

  21. Build a publicly controlled compute foundation for public-purpose models in areas of demonstrated Australian national need.

    The stronger strategy is therefore a publicly controlled compute foundation supporting public purpose models in areas where Australia has a demonstrated national need.

  22. Require evidence of each public AI model’s need/usefulness; determine ownership and durable public rights before Commonwealth funding, compute or strategic datasets are supplied, preventing automatic private enclosure of publicly funded knowledge.

    Individual projects should still require evidence of need and usefulness rather than assuming that every public activity requires a bespoke AI model. Where the Commonwealth funds development, supplies the compute or contributes strategic public datasets, ownership and durable public rights should be determined from the beginning. Publicly funded knowledge should not automatically become privately enclosed intellectual property that government later pays to access.

  23. Give every high-consequence Commonwealth AI system a named accountable public officer and a human decision pathway with genuine authority to examine evidence, correct errors and overturn automated or AI-assisted outcomes.

    Every high consequence system should have a named accountable public officer and a human decision pathway with genuine authority to examine evidence, correct errors and overturn an automated or AI assisted outcome.

  24. Give high-consequence Commonwealth AI a documented lawful purpose, impact assessment, defined data sources, appropriate pre-deployment testing, logging, monitoring, incident procedures and meaningful review.

    Such systems should have a documented lawful purpose, impact assessment, defined data sources, appropriate pre deployment testing, logging, continuing monitoring, incident procedures and meaningful review.

  25. Ensure a person materially affected by AI knows when it played a significant role, receives intelligible reasons and retains the same substantive review rights as if the decision had been made without AI.

    A person materially affected by AI should know when it played a significant role, receive intelligible reasons and retain the same substantive review rights that would exist if the decision had been made without AI.

  26. Establish a public register of material Commonwealth AI systems (purpose, responsible agency, technology, data categories, human oversight, assessment status and material incidents), subject to narrowly defined national security exceptions.

    A public register should identify material Commonwealth AI systems, their purpose, responsible agency, broad technology, significant data categories, human oversight arrangements, assessment status and material incidents, subject to narrowly defined national security exceptions.

  27. Renew assurance when model changes materially alter risk or capability.

    Model changes that materially alter risk or capability should also trigger renewed assurance rather than allowing initial approval to become permanent.

  28. Require subcontractor transparency in Commonwealth AI procurement.

    Contracts should preserve audit rights, necessary technical information, logs, data portability, interoperability, incident information, subcontractor transparency and practical exit arrangements.

  29. Ensure agencies know sensitive-data processing locations/access, material model-change processes and foreign-government, platform or supplier risks to continued operation.

    Agencies should know where sensitive information is processed, who can access it, how material model changes are made and whether a foreign government, upstream platform or supplier decision could prevent continued Australian operation.

  30. Do not let lowest initial contract price outweigh public expertise, switching capability, security, resilience and long-term bargaining power.

    Lowest initial contract price should not outweigh public expertise, switching capability, security, resilience and long-term bargaining power.

  31. Provide secure public technical careers with professional development, research links and continuity; expand university/TAFE capacity where shortages exist, supported by educators, laboratories, equipment, supervisors and work opportunities.

    Public service technical careers should provide meaningful work, professional development, research links and enough continuity to retain institutional knowledge. University and TAFE capacity should expand where genuine shortages exist, but enrolment targets alone are insufficient. Additional places require educators, laboratories, equipment, supervisors and practical work opportunities, while regional training should be connected to real regional capability rather than created as an isolated numerical target.

  32. Distinguish genuinely non-commercial, bounded, transparent and lawful public-interest research from unrestricted commercial exploitation in copyright arrangements.

    Public interest research should nevertheless be distinguished from unrestricted commercial exploitation because the public interest can justify different arrangements where use is genuinely non-commercial, bounded, transparent and lawful.

  33. Respect relevant First Nations rights holders and communities in cultural-material governance; Australian origin must not imply model-training permission.

    First Nations cultural material requires particular protection because technological ability to reproduce or infer cultural knowledge does not itself create legitimate authority to exploit it. Governance should respect relevant rights holders and communities rather than treat Australian origin as automatic permission for model training.

  34. Minimise age/identity-assurance data, avoid unnecessary central repositories and retain information only as long as genuinely required.

    Age and identity assurance should therefore minimise data collection, avoid unnecessary central repositories and retain information only for as long as genuinely required.

  35. Apply least privilege, segmentation, strong authentication, logging, controlled third-party access, recoverable backups and rapid isolation to important AI systems.

    The secure by default principle developed in Cyber security for small to medium sized businesses and organisations should apply to important AI systems through least privilege, segmentation, strong authentication, logging, controlled third party access, recoverable backups and rapid isolation.

  36. Retain fallback modes and domestic technical knowledge for high-consequence Commonwealth/critical-infrastructure functions to continue without foreign black-box dependence.

    High consequence Commonwealth and critical infrastructure functions should therefore retain fallback operating modes and enough domestic technical knowledge to continue essential activity without continuous dependence upon a foreign controlled black box.

  37. Assess critical compute and major data centres for ownership/control, Australian data access, foreign legal exposure, supply-chain dependence, continuity, infrastructure requirements, public assistance and retained Australian value.

    Critical compute and major data centres should therefore be assessed for ownership, control, Australian data access, foreign legal exposure, supply chain dependence, continuity, infrastructure requirements, public assistance and the value likely to remain in Australia.

  38. Have competition policy examine infrastructure control, switching, interoperability and dependency rather than only the retail price of individual AI products.

    Competition policy should consequently examine infrastructure control, switching, interoperability and dependency rather than only the retail price of individual AI products.

  39. Enable practical data/operation switching between providers without prohibitive technical or contractual barriers.

    Australian businesses should be able to move data and operations between providers without prohibitive technical or contractual barriers where practical.

  40. Distinguish legal, enforcement and technical capability gaps before deciding on legislation; retest the 2024 legislative architecture rather than treating it as settled.

    Australia should not create a vast new body of law merely because a product contains AI because many harmful activities already fall within consumer, competition, privacy, corporations, employment, administrative, safety, criminal and other laws. The inquiry should distinguish legal gaps from enforcement gaps and technical capability gaps before deciding which new legislation is required. The 2024 Committee recommended dedicated whole of economy legislation for high- risk AI while also recognising that substantial existing regulation already applied and that duplication was a possible disadvantage. The present inquiry should therefore retest the appropriate legislative architecture rather than treating the earlier recommendation as a settled conclusion.

  41. Consider an AI Act for cross-sector risks existing laws cannot handle coherently, while addressing enforcement, capture, expertise and institutional-responsibility problems; identify the problem before choosing legislation.

    A dedicated AI Act may ultimately be justified for cross sector risks that existing legislation cannot handle coherently, but another statute will not itself correct weak enforcement, regulatory capture, poor technical expertise or unclear institutional responsibility. Parliament should determine the problem first and the legislative vehicle second.

  42. Base new duties on consequence, practical control and risk; impose substantially stronger obligations on systems affecting liberty, income, health, employment, essential services, government, infrastructure or large populations than on low-consequence tools.

    New duties should therefore be based primarily upon consequence, practical control and risk rather than the mere presence of AI. Systems affecting liberty, income, health, employment, essential services, public administration, critical infrastructure or large populations should face substantially stronger obligations than low consequence productivity tools.

  43. Publish a regulatory grid naming the institution with primary responsibility and powers for each major risk, cross-agency escalation routes and legislative gaps.

    A public regulatory grid should identify which institution carries primary responsibility for each major risk, what powers it possesses, how cross agency escalation occurs and where legislative gaps remain.

  44. Have AISI supply technical evaluation, horizon scanning, testing methods and regulator support; manage independence, conflicts and revolving-door risks, publishing significant public-interest findings unless genuine security requires withholding.

    The Australian AI Safety Institute should provide specialist technical evaluation, horizon scanning, testing methodologies and cross regulator support rather than become a repository into which existing regulators transfer their substantive responsibilities. Its independence, conflicts of interest and revolving door risks should be actively managed, while significant public interest findings should be published unless genuine security requirements prevent disclosure.

  45. Give AISI and relevant regulators access to specialist technical staff.

    The Institute and relevant regulators should have meaningful access to sovereign public compute and specialist technical staff.

  46. For high-consequence systems, enable regulators to obtain information, require material-incident reports, commission independent assessment and compel remediation of legal breaches; capture material near misses/model changes revealing systemic risk.

    For high consequence systems, responsible regulators should be able to obtain necessary information, require material incident reporting, commission independent technical assessment and compel remediation where legal duties are breached. Material near misses and significant model changes should also be captured where they reveal systemic risk rather than waiting until actual injury occurs.

  47. Require context-specific pre-deployment testing and continuing assurance responsive to models, data, users and threats; require independent red teaming/evaluation where consequences justify the burden.

    Testing before deployment should reflect the real context of use, while continuing assurance should account for changing models, data, users and threat environments. Independent red teaming and technical evaluation should be required where potential consequences justify the burden.

  48. Maintain viable critical-system fallback arrangements that do not assume uninterrupted cloud, network, identity-platform or external-model availability or concentrate fallback on the same provider.

    Critical systems should maintain viable fallback arrangements that do not assume uninterrupted availability of every cloud service, network, identity platform and external model. Concentrating all fallback arrangements on the same underlying provider is not resilience.

  49. Align connection, energy and infrastructure rules with real system costs.

    Connection, energy and infrastructure rules should therefore align private decisions with the real system costs they create.

  50. Have the Commonwealth fund, develop, own, retain, operate and renew independently operable sovereign AI/data centres, with public compute, secure data, expertise and continuity for essential public, scientific, regulatory, infrastructure and security functions.

    The Committee should recommend that the Commonwealth fund, develop, own, retain, operate and continually renew a sovereign standalone Australian AI and data centre capability as essential national public infrastructure. The capability should be independently operable under Australian authority and contain sufficient publicly controlled compute, secure data infrastructure, technical expertise, resilience and continuity arrangements to maintain essential public, scientific, regulatory, critical infrastructure and national security functions without exclusive dependence upon a foreign hyperscaler or private provider.

  51. Make the sovereign public capability the dependable foundation of Australia's AI system rather than a residual market-failure mechanism.

    The Committee should recommend that the sovereign public capability form the dependable foundation of Australia's AI system rather than a residual market failure mechanism.

  52. Have private Australian and foreign providers supplement and compete with the sovereign public foundation, while strategically essential functions remain capable of public operation where commercial or geopolitical conditions change.

    Private Australian and foreign providers should supplement and compete with that foundation, while strategically essential functions remain capable of public operation where commercial or geopolitical conditions change.

  53. Treat the 2024 Select Committee report as an important starting point rather than a completed settlement.

    The Committee should recommend that the 2024 Select Committee report be treated as an important starting point rather than a completed settlement.

  54. Strengthen the 2024 sovereign capability recommendation by requiring the Commonwealth to retain sufficient operational capability, and make its national interest infrastructure recommendation measurable through ownership, control, continuity, resource additionality and public value tests.

    Its sovereign capability recommendation should be strengthened by requiring the Commonwealth itself to retain sufficient operational capability, while its national interest infrastructure recommendation should be made measurable through ownership, control, continuity, resource additionality and public value tests.

  55. Judge AI investment/adoption against productive capacity, living standards, sovereignty, resilience, distribution, accountability and resource use; recognise Commonwealth monetary sovereignty alongside real-resource, inflation, external-supply and ecological constraints.

    The Committee should recommend an explicit good public purpose framework under which AI investment and adoption are judged against useful productive capacity, living standards, sovereignty, resilience, distribution, accountability and real resource use. That framework should recognise Commonwealth monetary sovereignty while expressly acknowledging real resource, inflation, external supply and ecological constraints.

  56. Make sovereign capability geographically/technically resilient, interoperable where practicable, with diverse supply/fallback; govern through a statutory public-interest mandate, independent technical assurance, parliamentary reporting, transparent procurement and strategic public-asset preservation.

    The Committee should recommend that the sovereign public capability be geographically and technically resilient, use interoperable architectures where practicable and maintain diversified supply and fallback arrangements. Governance should include a statutory public interest mandate, independent technical assurance, parliamentary reporting, transparent procurement and preservation of strategic public assets.

  57. Default sensitive compulsory Commonwealth information to publicly controlled infrastructure and high-consequence Commonwealth AI to sovereign public infrastructure where capable, with documented justification for departures.

    The Committee should recommend that sensitive compulsory Commonwealth information default to publicly controlled infrastructure and that high consequence Commonwealth AI default to sovereign public infrastructure where the Commonwealth possesses the necessary technical capability. Departures should require a documented justification

  58. Develop a national AI/data-centre resource plan covering electricity, firming, transmission, water, land, workforce, training, materials, semiconductors, communications and strategic supply chains.

    The Committee should recommend a national real resource plan for AI and data centre development covering electricity, firming, transmission, water, land, workforce, training, materials, semiconductors, communications and strategic supply chains.

  59. Assess major projects’ additionality and opportunity cost alongside housing, transport, agriculture, manufacturing, health, energy and other uses of the same resources.

    Major projects should be assessed for additionality and opportunity cost alongside housing, transport, agriculture, manufacturing, health, energy and other claims upon the same resources.

  60. Have major private data centres bear attributable connection and augmentation costs and provide or underwrite additional capacity where appropriate, rather than socialising new infrastructure demands.

    The Committee should recommend that major private data centres bear attributable connection and augmentation costs and provide or underwrite additional capacity where appropriate rather than simply socialising new infrastructure demands.

  61. Secure enduring public value appropriate to each project from material Commonwealth support for private AI infrastructure: ownership/equity, compute/access/IP rights, research capacity, repayment or other measurable benefits.

    The Committee should recommend that material Commonwealth support for private AI infrastructure purchase identifiable enduring public value rather than merely de risk private ownership. Public value may include ownership, equity, compute rights, infrastructure access, intellectual property rights, research capacity, repayment or other measurable benefits appropriate to the project.

  62. Expand sovereign AI skills through public technical careers, apprenticeships, paid training, scholarships, university/TAFE capacity and research roles; build capacity rather than drawing specialists away from housing, energy, infrastructure, agriculture and other constrained sectors.

    The Committee should recommend deliberate expansion of the real workforce needed for sovereign AI through public technical careers, apprenticeships, paid training, scholarships, university and TAFE capacity and research positions. Workforce policy should expand national capability rather than simply bidding engineers, electricians and technical specialists away from housing, energy, infrastructure, agriculture and other constrained sectors.

  63. Make responsibility for AI risk follow practical control, information and capacity so developers, platforms, infrastructure providers and major deployers cannot transfer controlled risks to consumers, workers and SMEs through disclaimers or standard form contracts.

    The Committee should recommend that responsibility for AI risk follow practical control, information and capacity so that developers, platforms, infrastructure providers and major deployers cannot transfer risks they control onto consumers, workers and SMEs through disclaimers or standard form contracts.

  64. Keep regulatory compliance proportionate and avoid turning smaller businesses into auditors of systems they cannot inspect.

    Regulatory compliance should remain proportionate and should not turn smaller businesses into auditors of systems they cannot inspect.

  65. Have Commonwealth procurement preserve data control, technical information, audit rights, logs, interoperability, portability, incident information, continuity and exit arrangements.

    The Committee should recommend that Commonwealth procurement preserve data control, technical information, audit rights, system logs, interoperability, portability, incident information, continuity and practical exit arrangements.

  66. Have procurement supplement rather than substitute for sovereign capability, and treat vendor lock-in as a strategic cost.

    Procurement should supplement sovereign capability rather than substitute for it, and vendor lock in should be treated as a strategic cost.

  67. Protect workers from involuntary unemployment being the adjustment mechanism, with paid retraining, secure transition pathways and consideration of a federally funded, locally administered Job Guarantee.

    The Committee should recommend that AI productivity policy protect workers from involuntary unemployment becoming the adjustment mechanism for useful technological change. Paid retraining and secure transition pathways should be combined with consideration of a federally funded and locally administered Job Guarantee

  68. Require commercial AI developers to provide sufficient transparency about protected Australian material for copyright to be enforceable.

    The Committee should recommend that commercial AI developers provide sufficient transparency concerning protected Australian material for copyright rights to be enforceable

  69. Develop workable licensing and remuneration where commercial models derive material value from protected works.

    and that workable licensing and remuneration arrangements be developed where commercial models derive material value from protected works.

  70. Concentrate scam, deepfake and synthetic identity responsibility at the banking, telecommunications, platform, identity and AI points where harm can practically be prevented or contained.

    The Committee should recommend that scam, deepfake and synthetic identity policy concentrate responsibility at the banking, telecommunications, platform, identity and AI system points where harm can practically be prevented or contained.

  71. Avoid converting protection of children and vulnerable Australians into unnecessary universal surveillance or mass identity collection, and keep human and non-digital pathways for essential public and financial services.

    Protection of children and vulnerable Australians should not be converted into unnecessary universal surveillance or mass identity collection, while essential public and financial services should retain meaningful human and non-digital pathways.

  72. Have the Australian AI Safety Institute provide independent technical expertise and cross-regulator support, including access to sovereign public compute, while existing specialist regulators retain clear responsibility within their legal domains.

    The Australian AI Safety Institute should provide independent technical expertise and cross regulator support, including access to sovereign public compute, while existing specialist regulators retain clear responsibility within their legal domains.

  73. Require risk-proportionate pre-deployment testing, continuing monitoring, material incident reporting, independent evaluation and continuity planning for high-consequence systems.

    The Committee should recommend risk proportionate pre deployment testing, continuing monitoring, material incident reporting, independent evaluation and continuity planning for high consequence systems.

  74. Ensure critical government and infrastructure functions do not depend exclusively on technology Australia cannot adequately understand, audit, replace or continue operating when an external provider becomes unavailable.

    Critical government and infrastructure functions should not depend exclusively upon technology Australia cannot adequately understand, audit, replace or continue operating when an external provider becomes unavailable.

  75. Assess foreign AI investment by the enduring capability, public value and resource demands it leaves in Australia rather than headline financial investment alone.

    The Committee should recommend that foreign AI investment be assessed by the enduring capability, public value and resource demands it leaves in Australia rather than headline financial investment alone.

  76. Welcome productive foreign participation while retaining Commonwealth capacity for public alternatives where strategic dependency, excessive rent extraction or loss of control would otherwise result.

    Foreign participation should remain welcome where productive, but the Commonwealth should retain the capacity to build and operate public alternatives where strategic dependency, excessive rent extraction or loss of control would otherwise result.

Read the original

Open in the interactive guide