Submission 19
UNSW AI Institute
Submitter group: Universities.
An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.
Summary
Read the original submission for the full text.
Argues that Australia needs a funded national AI strategy and sovereign capability rather than reliance on data-centre investment alone. Proposes reserved public-interest compute, local R&D obligations on major AI firms, data sovereignty and copyright protections, worker participation, a better-resourced AI Safety Institute and risk-based duties on high-impact AI.
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.
Argues that public investment and policy aligned to a National AI Strategy that builds Australia's independent AI capability is the single most important issue for the Committee.
Public investment and policy aligned to a National AI Strategy that builds Australia’s independent AI capability is the single most important issue this Committee can address.
A program corrected obvious text-recognition errors in this quote for display.
States that Australia's institutional GPU base is around 2,200 accelerators, most several hardware generations old.
Australia's entire institutional GPU base across every university, CSIRO and national facility combined is around 2,200 accelerators, most of them several hardware generations old.
Argues that building data centres does not necessarily increase Australian AI capability, and says that Expectation 5 for data centre developers requires access to compute on 'favourable terms' for Australian entities.
Building data centres in Australia does not necessarily increase our AI capability. Expectation 5 for Data Centre and AI Infrastructure Developers requires access to compute on “favourable terms” for Australian entities.
A program corrected obvious text-recognition errors in this quote for display.
Says that public datasets, taxpayer-funded research and cultural material may be open for public benefit, but that this does not entitle every commercial training use to unrestricted access.
Public datasets, taxpayer-funded research and cultural material may be open for public benefit, but that does not entitle every commercial training use to unrestricted access.
Argues that Australia should not weaken copyright in the hope of attracting data-centre investment.
Australia should not weaken copyright in the hope of attracting data-centre investment. A stable rights framework is an asset because it enables licensing, investment and trust.
Argues that existing law is fragmented and does not consistently require disclosure, traceability, independent testing or a route to contest AI-assisted outcomes.
Existing law is fragmented across privacy, consumer protection, online safety, anti-discrimination and administrative law, which do not consistently require disclosure, traceability, independent testing or a practical route to contest AI-assisted outcomes.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Develop a national AI strategy with an ambition to build Australian AI capability and a plan to achieve it.
Develop an AI Strategy for Australia that sets an ambition to build our AI capability and a plan to achieve it.
A program corrected obvious text-recognition errors in this quote for display.
Treat digital infrastructure as critical national infrastructure for budget, planning, energy, water and research purposes.
Treat digital infrastructure as critical national infrastructure for budget, planning, energy, water and research purposes, as it is already treated as critical for cybersecurity under the SOCI Act.
Establish a funded public-interest compute base and a dedicated AI research funding vehicle in the forward estimates.
Establish a funded public-interest compute base and a dedicated AI research funding vehicle in the forward estimates.
A program corrected obvious text-recognition errors in this quote for display.
Create an AI/science compact requiring training-data provenance, substantial-contribution attribution, relevant evaluations, confidential and unfinished research protection, and fair returns when publicly funded work creates commercial value.
The response should be a national compact for AI and science. Public funding agreements should require provenance for training data, attribution for substantial scientific contributions, access to relevant evaluation results, protection for confidential and unfinished research, and a fair return when publicly funded work creates commercial value.
A program corrected obvious text-recognition errors in this quote for display.
Support open models and open datasets where security, privacy and Indigenous governance permit, and fund accessible public compute.
The compact should support open models and open datasets where security, privacy and Indigenous governance permit, and should fund accessible public compute.
Require data stewardship and public-interest access, alongside provenance, attribution and public returns, when publicly funded research materially contributes to commercial AI systems.
Protect the conditions for open scientific discovery. Require provenance, attribution, data stewardship, public-interest access and a return on public investment when publicly funded research materially contributes to commercial AI systems.
A program corrected obvious text-recognition errors in this quote for display.
Require new private-sector contributions to strengthen rather than displace sovereign research infrastructure and public investment.
New private-sector contributions must strengthen, not displace, existing sovereign research infrastructure and public investment.
Protect researchers through clear IP ownership, secure data handling and safeguards against inappropriate reuse of research assets or models.
Clear IP ownership, secure data handling and safeguards against inappropriate reuse of research assets or models.
Require clear reporting, benchmarking and accountability for promised and delivered capacity and research utilisation.
Clear reporting, benchmarking and accountability for capacity promised, capacity delivered and research utilisation.
Coordinate access through established national research infrastructure mechanisms rather than ad hoc new structures.
Access should be coordinated through established national research infrastructure mechanisms rather than ad hoc new structures.
Require no vendor lock-in, researcher portability and interoperability.
No vendor lock-in. Researchers need portability, interoperability
Require stable pricing and reliable access for researchers across multi-year grant and research lifecycles.
Stable pricing and reliable access across multi-year grant and research lifecycles.
Review compute arrangements at least annually, because AI infrastructure, architectures and capability requirements change rapidly.
Arrangements should be reviewed at least annually because AI infrastructure, architectures and capability requirements rapidly change.
A program corrected obvious text-recognition errors in this quote for display.
Make reserved compute, independent advice for regional SMEs, shared testbeds and training available beyond capital cities through regional universities, TAFEs and industry networks, with reporting on local outcomes.
Benefits should also be distributed beyond capital-city data-centre precincts. Reserved compute, independent advice for regional SMEs, shared testbeds and training should be available through regional universities, TAFEs and industry networks, with reporting on local jobs, skills and supply-chain outcomes.
Condition Commonwealth/state facilitation on operators and owners reserving verifiable compute for Australian researchers, start-ups, SMEs, family businesses, regional universities and public-interest users; permit in-kind/tradeable discharge through one national research-infrastructure counterparty and pool.
Require, as a binding condition of Commonwealth or state facilitation, that AI infrastructure operators and owners reserve a defined and independently verifiable share of installed compute for Australian researchers, start-ups, SMEs, family businesses, regional universities and public-interest users, discharged either in kind or through a tradeable obligation to another provider, and administered through a single national counterparty drawing on Australia's existing Tier 1 and Tier 2 research infrastructure providers so that entitlements aggregate into one usable national pool rather than fragmenting across sites and vendors.
A program corrected obvious text-recognition errors in this quote for display.
Require major AI companies’ local R&D, reform the large-company R&D Tax Incentive, retain generated IP in Australia and create rural-RDC-style AI research funding; apply obligations to compute controllers with major-tenant contractual flow-down.
Adopt a mandatory local AI research and development presence requirement for major AI companies operating in Australia, reform the R&D Tax Incentive so that it remains effective for large companies, require that intellectual property generated remains in Australia, and establish a dedicated AI research funding mechanism on the model of the rural research and development corporations. Apply these obligations to the party that controls the compute, with contractual flow-down to major tenant agreements.
A program corrected obvious text-recognition errors in this quote for display.
Build the public data layer around provenance, consent, purpose limitation, security and benefit sharing.
The public data layer should be built around provenance, consent, purpose limitation, security and benefit sharing.
Apply strict public-interest criteria to training-data use: attribution, access to evaluations, ongoing remuneration, Australian hosting for sensitive data and return of derived datasets and methods.
Prioritising Australian AI models and applying strict public-interest criteria on attribution, access to resulting evaluations, ongoing remuneration, Australian hosting for sensitive data, and right of return of derived datasets and methods.
Build an Australian data layer based on consent, provenance, Indigenous Data Sovereignty, the CARE principles and the public interest.
Build an Australian data layer based on consent, provenance, Indigenous Data Sovereignty, the CARE principles and the public-interest.
Ideally use publicly funded data in Australian AI models rather than transferring it into foreign-controlled models without a governed agreement covering attribution, ongoing return and benefit sharing.
Publicly funded data should ideally be used in Australian AI models, not transferred into foreign-controlled models without a governed agreement covering attribution, ongoing return and appropriate benefit sharing.
Use Australian machine-unlearning expertise when developing verifiable erasure and deletion requirements.
Australian researchers have internationally recognised expertise in machine unlearning, which should be used in developing these requirements.
Increase the funding, technical resources and authority of the Australian AI Safety Institute in line with comparable international bodies.
Increase the funding, technical resources and authority of the Australian AI Safety Institute in line with comparable international bodies.
Give the AI Safety Institute access to models, compute, data and controlled environments to independently evaluate and certify high-risk AI systems before deployment, with formal access to Australian academic expertise.
Increase the funding, technical resources and authority of the Australian AI Safety Institute in line with comparable international bodies. Give it access to the models, compute, data and controlled environments needed to independently evaluate and certify high-risk AI systems before deployment, supported by formal access to Australian academic expertise.
Require incident reporting, independent evaluation, red-teaming and supply-chain and foreign-interference assessment for AI used in critical settings.
Require incident reporting, independent evaluation, red-teaming and supply-chain and foreign-interference assessment for AI used in critical settings.
Invest in domestic AI capability and cybersecurity skills.
Invest in domestic AI capability and cybersecurity skills
Identify foreign ownership, model-update control, remote administration, data export, lock-in, hardware dependencies, critical spare parts, cyber-incident pathways and provider withdrawal consequences in AI procurement and planning.
Procurement and planning should identify foreign ownership, control of model updates, remote administration, data export, vendor lock-in, hardware dependencies, critical spare parts, cyber-incident pathways and the consequences of a provider withdrawing its service.
Base any data-erasure requirements on demonstrated machine-unlearning capability.
base any data-erasure requirements on demonstrated machine-unlearning capability
Pursue international agreement on meaningful human control of military AI, including excluding AI from nuclear weapons command, control and launch decisions.
pursue international agreement on meaningful human control of military AI, including the exclusion of AI from nuclear weapons command, control and launch decisions.
Provide shared testbeds so small manufacturers can trial robotics automation before committing capital.
Australian robotics companies would benefit from robust sovereign procurement preferences to match the scale of the recently expanded Defence Integrated Investment Program, and shared testbeds that let small manufacturers trial automation before committing capital.
Design physical-AI adoption initiatives such as Singapore’s vouchers in consultation with unions.
The instant asset write-off, now permanently set at $20,000, is too low to capture most robotics purchases and should be reviewed, while further encouraging adoption of physical AI via successful overseas policy initiatives such as Singapore’s voucher system, designed in consultation with unions, would have demonstrable benefits.
Reduce barriers to capital for robotics companies through vehicles such as the National Reconstruction Fund.
Reduce barriers to capital for robotics companies through vehicles such as the National Reconstruction Fund
Review the instant asset write-off threshold for robotics and automation equipment.
review the instant asset write-off threshold for robotics and automation equipment
Consider an adoption-voucher scheme of the kind Singapore introduced.
consider an adoption-voucher scheme of the kind Singapore introduced
Extend elements of the sovereign procurement preference applied to defence uncrewed systems to civilian robotics contracts.
extend elements of the sovereign procurement preference already applied to defence uncrewed systems to civilian robotics contracts.
Develop industry policy to support the scaling of Australian AI and robotics companies.
Develop industry policy to support the scaling of Australian AI and robotics companies.
Establish a national register of strategic technology transactions drawing on FIRB, ATO and ASIC data.
Establish a national register of strategic technology transactions drawing on FIRB, ATO and ASIC data
Treat AI, industrial automation and critical technology software as sensitive for foreign-investment screening, with a public-benefit test and enforceable undertakings on Australian IP holding, data residency, minimum local R&D spend and headcount.
treat AI, industrial automation and critical technology software as sensitive for foreign-investment screening, with a public- benefit test and enforceable undertakings on Australian IP holding, data residency, minimum local R&D spend and headcount
Establish a patient co-investment growth fund for companies in the $10m to $200m revenue band, including finance for outbound acquisitions by Australian companies.
establish a patient co-investment growth fund focused on companies in the $10m to $200m revenue band, including finance for outbound acquisitions by Australian companies to reverse the trend.
Give workers a meaningful role in decisions on AI deployment, work design, surveillance, performance assessment, safety and retraining, with consultation starting before procurement.
Workers should have a meaningful role in decisions about deployment, work design, surveillance, performance assessment, safety and retraining, and consultation should start before procurement.
Require public procurement to show that workers were involved, risks were assessed and affected people have a route to challenge an automated decision.
Public procurement should require evidence that workers were involved, that risks were assessed, and that affected people have a route to challenge an automated decision.
Require AI infrastructure developments to support apprenticeships, local training partnerships, regional and TAFE pathways, industry-linked PhDs and research–industry interchange.
Data centre and AI infrastructure developments should be required to support a range of AI capability uplift programs including apprenticeships, local training partnerships, regional and TAFE pathways, industry-linked PhD placements and programs that encourage interchange between research and industry.
Coordinate national workforce planning in consultation with workers and unions, with transition and retraining plans, worker input into deployment, safe and secure jobs, and education pathways from TAFE and apprenticeships through research to advanced engineering.
Coordinated national workforce planning in consultation with workers and unions, transition and retraining plans, worker input into deployment, safe and secure jobs, and education pathways running from TAFE and apprenticeships through to research and advanced engineering.
Coordinate the workforce work nationally through a Sovereign AI Coordination Vehicle, advised by a tripartite body representing government, employers and workers.
This work should be coordinated nationally by a Sovereign AI Coordination Vehicle, advised by a tripartite body representing government, employers and workers.
Attach direct workforce commitments to AI infrastructure development, with evidence of actual training and employment outcomes beyond vendor-specific training modules.
Direct workforce commitments should attach to AI infrastructure development, as well as evidence of actual training and employment outcomes beyond supply of vendor-specific training modules.
Maintain existing copyright protections and do not introduce a broad exception for commercial AI training.
Maintain Australia’s existing copyright protections and do not introduce a broad exception for commercial AI training.
Require AI developers to keep auditable records and disclose sufficient information about the Australian creative, cultural, media and publicly funded research material used to train their models.
Require AI developers to keep auditable records and disclose sufficient information about the Australian creative, cultural, media and publicly funded research material used to train their models.
Developers should obtain licences directly or through collective arrangements and provide ongoing remuneration where material is used to generate commercial value.
Developers should obtain licences directly or through collective arrangements and provide ongoing remuneration where this material is used to generate commercial value.
Any copyright exception should be narrow and transparent and preserve provenance, ethical conditions and rights to correction, withdrawal and compensation.
Any exception should be narrow and transparent and should preserve provenance, ethical conditions and rights to correction, withdrawal and compensation.
Establish separate, enforceable protections for Indigenous Cultural and Intellectual Property, developed with Aboriginal and Torres Strait Islander peoples.
Establish separate, enforceable protections for Indigenous Cultural and Intellectual Property, developed with Aboriginal and Torres Strait Islander peoples.
Require high-impact AI providers to disclose material changes, maintain logs, support independent testing and cooperate with independent external incident investigations.
providers of high-impact systems should disclose material changes, maintain logs, support independent testing and cooperate with independent external incident investigations rather than supplying their own analysis.
Through the Office of AI, align the proposed Australian AI standards with consumer protection, privacy, online safety and digital duty of care reforms.
Through the Office of AI, align the proposed Australian AI standards with consumer protection, privacy, online safety and digital duty of care reforms.
Place enforceable, risk-based duties on providers and deployers of high-impact AI, prioritising scams, deepfakes, companion chatbots and agents connected to payments, health records or public benefits.
The resulting framework should place enforceable, risk-based duties on providers and deployers of high-impact AI, with priority given to scams, deepfakes, companion chatbots and agents connected to payments, health records or public benefits.
Require content provenance, accessible reporting, redress, independent testing and cooperation with external incident investigations.
Require content provenance, accessible reporting, redress, independent testing and cooperation with external incident investigations.
Have the AI Safety Institute develop common tests and incident classifications, drawing on Australian research in anomaly detection and other safety-relevant fields.
The Australian AI Safety Institute should develop common tests and incident classifications, drawing on Australian research in anomaly detection and other safety-relevant fields.
Read the original
In some quotes a program corrected obvious text-recognition errors for display.