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 199

Kingston AI Group

Submitter group: Research institutes.

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 for investment to retain Australian AI research and build sovereign capability through mandated local R&D, talent and domain-specific models. Says its regulatory expertise is narrower and defers on child safety. Appendix A separately proposes training-related copyright reform, streamlined approvals, a data-centre levy, reserved compute and renewable-generation conditions; the main submission distinguishes academic research from commercial content exploitation.

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 Australia’s AI constraint is scale, infrastructure and coordination to convert talent into sovereign economic output, rather than a talent shortage.

    Australia's constraint on capturing AI-driven productivity and competitiveness gains is not a shortage of talent. It is the absence of the scale, infrastructure and coordination needed to convert that talent into sovereign economic output, precisely the gap a National AI strategy should be designed to close.

  2. Argues that sovereign capability is tested by what Australia could still do if access were withdrawn, for example by a supplier revising its terms or a foreign export control directive.

    The test is what Australia could still do if access were withdrawn. Capability that stops working when a supplier revises its terms, or a foreign government issues an export control directive, was never sovereign.

  3. Proposes that a local AI R&D presence obligation be attached to AI training operators that develop and control frontier models, rather than to data-centre operators.

    This local R&D presence obligation should be attached to AI training operators - the entities that develop and control frontier models - rather than to data-centre operators, who host but do not own the compute, models or IP that determine where value is captured.

  4. Argues Australia need not compete at frontier scale to address offshore-dependence risks, but needs domain-specific models trained on Australian data in sectors of existing expertise.

    Australia does not need to compete at frontier scale to address this risk. It needs domain-specific models, trained on Australian data, in the sectors where Australian data and Australian expertise already exist.

  5. Argues that regulation drafted without reference to evidence on machine unlearning risks unenforceable obligations or costly workarounds such as full model retraining.

    Regulation drafted without reference to this evidence risks setting obligations that are either unenforceable in practice or achievable only through crude and costly workarounds, such as full model retraining, that a mature machine-unlearning capability would avoid.

  6. Argues that Australia's workforce priority in AI should be retaining PhD-level, deep-tech research talent through research career pathways, HDR scholarships and corporate R&D placements.

    Australia's workforce priority in AI should be retaining its PhD-level, deep-tech research talent, cutting-edge expertise that underpins every recommendation in this submission, by funding research career pathways, HDR scholarships and corporate R&D placements, rather than letting broader displacement and adoption debates crowd out this more immediate retention problem.

Recommendations

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

  1. Invest to convert Australia's world-standard AI research capability into sovereign economic output, as deliberate industrial strategy rather than assuming adoption of overseas AI products will deliver equivalent gains.

    Australia should invest to convert its demonstrated, world-standard AI research capability into sovereign economic output, treating this as a deliberate act of industrial strategy rather than an assumption that adoption of overseas AI products will deliver equivalent productivity, competitiveness and gains in living-standards.

  2. Require any mandated or incentivised local AI R&D presence to be distributed across states and regions, and fund domain-specific AI applications in priority areas such as health, agriculture and resources extending to regional and remote communities.

    Any mandated or incentivised local AI R&D presence should be required to be distributed across states and regions, not concentrated in capital cities, and Commonwealth AI investment should deliberately fund domain-specific AI applications in priority areas such as health, agriculture and resources, that extend specialist capability into regional and remote communities.

  3. Fund narrow, domain-specific AI models in sectors of comparative advantage (health and medical research, agriculture, resources, advanced manufacturing, defence), and address the tax and incentive settings that make it more attractive to locate substantial AI R&D offshore.

    Australia should fund the development of narrow, domain-specific AI models in sectors of comparative advantage such as health and medical research, agriculture, resources, advanced manufacturing and defence, building on the concentrations of research strength (embodied AI, trustworthy AI, clinical AI governance) already evident in Australia's research base, and should address the tax and incentive settings that currently make it more attractive for Australian companies to locate substantial AI R&D operations offshore.

  4. Extend the AI Adopt Program and similar schemes to shared, off-the-shelf sector-specific AI accessible to small, medium and family businesses, ensuring data security and sovereignty.

    Government should extend programs such as the AI Adopt Program to fund shared, off- the-shelf, sector-specific AI capability that small, medium and family businesses can license or access as a service rather than commission themselves, so the benefits of domain-specific AI research are not captured only by the large firms able to commission bespoke models and so that data security and sovereignty is assured.

  5. Require major model providers/frontier AI firms doing substantial Australian business to maintain long-term compulsory local R&D: roughly 10–15 PhD-level full-time-equivalent researchers distributed across states, rather than sales/deployment offices or optional/revenue-scaled commitments.

    Under this proposal, any major AI model provider or frontier AI company doing substantial business in Australia would be required to maintain a genuine local research and development capability, not a sales or deployment office, defined by an explicit floor of PhD-level, full-time-equivalent research staff (in the order of 10 to 15 people, benchmarked against the founding scale of early frontier AI labs), distributed across states rather than concentrated in capital cities, and sustained as a long-term, compulsory commitment rather than a revenue-scaled or optional arrangement.

    A program corrected obvious text-recognition errors in this quote for display.

  6. Pair the local R&D mandate with jointly company/government-funded hybrid 50/50 university–industry appointments, HDR scholarships and professor/researcher exchanges; offset industry costs through the R&D Tax Incentive and retain generated IP in Australia.

    We propose this be paired with hybrid 50/50 university-industry research appointments, higher-degree-by-research scholarships and professor/researcher exchange programs funded jointly by participating companies and government, use of Australia's existing R&D Tax Incentive to offset the cost to industry, and a requirement that IP generated through the mandated presence remains in Australia.

  7. Explicitly extend R&D target sectors beyond health, defence and agriculture to advanced manufacturing.

    We would extend the target sectors beyond those already emphasised in Australian innovation policy, such as health, defence and agriculture, to explicitly include advanced manufacturing.

  8. Back mandatory local AI R&D with an R&D Tax Incentive effective for large companies, Australian retention of generated IP, a dedicated large-scale computing fund and government venture capital with higher risk tolerance.

    Mandatory local AI R&D presence should be underpinned by matching incentives: reform of the R&D Tax Incentive so it remains effective for large companies, a requirement that IP generated through the mandated presence remains in Australia, a dedicated large-scale computing fund, and higher-risk- tolerant government venture capital to support it.

  9. Provide a dedicated computing fund of hundreds of millions of dollars for research competitiveness across AI, biotechnology, materials design and geology.

    Alongside this mandate, Australia needs a dedicated, large-scale computing fund, in the order of hundreds of millions of dollars, to underpin research competitiveness across AI, biotechnology, materials design and geology,

    A program corrected obvious text-recognition errors in this quote for display.

  10. Fund machine-unlearning, federated-learning and privacy-preserving infrastructure as a national data-sovereignty capability, to enable domain-specific models trained on Australian data in place of default dependence on offshore general-purpose models for sovereign-sensitive uses.

    Government should fund the development of machine-unlearning, federated-learning and privacy-preserving infrastructure as a national data-sovereignty capability, necessary to enable narrow, domain-specific AI models trained on Australian data in sectors such as health, agriculture, resources and defence, over continued default dependence on offshore-trained general-purpose models for sovereign- sensitive applications.

  11. Make retaining PhD-level deep-tech research talent the AI workforce priority, by funding research career pathways, HDR scholarships and corporate R&D placements.

    Australia's workforce priority in AI should be retaining its PhD-level, deep-tech research talent, cutting-edge expertise that underpins every recommendation in this submission, by funding research career pathways, HDR scholarships and corporate R&D placements, rather than letting broader displacement and adoption debates crowd out this more immediate retention problem.

  12. Draw Commonwealth AI transparency and accountability frameworks on domestic trustworthy and explainable AI research, resource government AI assurance functions to commission it, and use Commonwealth AI procurement as a lever to build local R&D capability.

    Commonwealth AI transparency and accountability frameworks should draw directly on Australia's domestic strength in trustworthy and explainable AI research, government AI assurance functions should be resourced to commission that expertise rather than relying only on offshore tools, and Commonwealth AI procurement should be used deliberately as a demand-side lever to build the local AI R&D capability.

  13. Where proposed AI regulation depends on technical feasibility, seek Australian AI/AI-safety researcher and KAIG expert input to ensure enforceable, technically realistic law.

    Where proposed AI regulation depends on technical feasibility (for example, data- erasure or model-correction obligations), the Committee should seek input from Australian AI and AI-safety researchers and KAIG experts on what is genuinely achievable, so that resulting law is both enforceable and technically realistic.

  14. Distinguish clearly in copyright and AI training law between non-commercial academic research use and commercial exploitation of Australian creative, cultural and media content, and keep IP generated using Australian data in Australia under any R&D mandate.

    Copyright and AI training law should distinguish clearly between non-commercial academic research use and commercial exploitation of Australian creative, cultural and media content, and any local AI R&D mandate adopted should require that IP generated using Australian data and content remains in Australia.

  15. Require critical AI infrastructure to meet demonstrated Australian security and compliance standards, including assurance against foreign-law disclosure obligations, and treat mandatory local R&D presence as a supply-chain resilience measure.

    Critical AI infrastructure should be required to meet demonstrated Australian security and compliance standards including assurance against disclosure obligations arising under foreign law; mandatory local AI R&D presence should be treated as a supply-chain resilience measure that reduces Australia's dependence on offshore-controlled AI infrastructure for sovereign-sensitive applications.

  16. Attach the local R&D presence obligation to AI training operators that develop and control frontier models, rather than to data-centre operators.

    This local R&D presence obligation should be attached to AI training operators - the entities that develop and control frontier models - rather than to data-centre operators, who host but do not own the compute, models or IP that determine where value is captured.

  17. Have the national security framework include an R&D presence requirement prioritising domains where Australia holds genuine research strength, rather than undifferentiated coverage of the full AI stack.

    The framework for national security should include R&D presence requirement which prioritises domains where Australia already holds genuine research strength, rather than pursuing undifferentiated coverage of the full AI stack (such as physical infrastructure, semiconductor production, cloud services to model training, deployment and application).

  18. Establish a standing, formal engagement mechanism between the AI Safety Institute and industry and research bodies with relevant technical depth, such as KAIG, so evaluation and security frameworks are calibrated to Australia's own risk profile.

    A standing, formal engagement mechanism should be established between Australia's AI Safety Institute and industry and research bodies with relevant technical depth, such as KAIG, to ensure evaluation and security frameworks are calibrated to Australia's own risk profile rather than developed in isolation from the capability needed to inform them.

  19. Consider formal mechanisms to embed academic AI-safety expertise in ongoing intelligence and security community engagement with AI risk.

    We would encourage the Committee to consider formal mechanisms for embedding this academic expertise into the Australian AI Safety Institute's advisory and technical assessment functions, and into intelligence and security community engagement with AI risk, on an ongoing rather than ad hoc basis, understanding that Australian bodies need to work in unison on a revolutionary critical technology era rather than in isolation.

  20. Consider the recommendations as a coherent national priority and package rather than a menu, particularly the mandatory local R&D presence.

    We urge the Committee to treat the following recommendations as a coherent package rather than a menu, in particular the mandatory local AI R&D presence proposed under ToR (e), which is the single measure most directly capable of converting Australia's demonstrated research strength, before the current window closes, into retained, sovereign, regionally distributed capability.

    A program corrected obvious text-recognition errors in this quote for display.

  21. Treat KAIG as a standing, evidence-generating resource for the inquiry and beyond.

    The Committee should treat KAIG as a standing, evidence-generating resource for the duration of this inquiry and beyond

  22. Appendix A (earlier Senate submission): enable model training on data without Australian copyright protection, separately treating Australian data and artists.

    1. Change copyright law to clearly enable AI model training on data that does not attract Australian copyright protection. A system providing a clear legal basis for training is a precondition for a domestic AI industry. The current law does not provide the support that artists require, yet it makes large-scale model training for export impossible.

  23. Appendix A: revisit a text-and-data-mining exception specifically supporting AI model training in Australia.

    Our submission is that an exception for text-and-data- mining be revisited, with the specific goal of supporting AI model training in Australia.

  24. Appendix A: create a coordinated, time-bound national approvals pathway for major data-centre and enabling-energy projects, with one entry point, concurrent assessments and statutory timeframes, preserving benefit-securing conditions.

    A coordinated, time-bound national approvals pathway for major data-centre and enabling-energy projects, and a single point of entry, concurrent rather than sequential assessment, and clear statutory timeframes, would materially improve Australia’s investment competitiveness. Importantly, streamlining the process is not the same as weakening the conditions attached to it. The benefit-securing conditions recommended in this submission can be built into a faster, clearer process.

  25. Appendix A: levy large-scale data centres for a dedicated long-term AI research, talent and capability fund; analyse levy base, rate and treatment of committed investment.

    A tax or levy on large-scale data-centre operations, with revenue directed to a dedicated long-term fund for AI research, talent and capability development, would convert a time-limited construction and operations cycle into a durable source of investment in domestic capability. The design of such a levy — base, rate, and treatment of investment already committed — warrants careful economic analysis, but the principle is well established.

    A program corrected obvious text-recognition errors in this quote for display.

  26. Appendix A: condition access to Australian land, power and public concessions on reserving a defined compute share for Australian universities, public-interest research, public agencies and start-ups.

    As a condition of access to Australian land, power and any public concession, a defined share of capacity should be reserved for Australian universities, public- interest research, public agencies and start-ups.

  27. Appendix A: condition grid connection on genuinely additional renewable generation and firming, using predictable data-centre demand to finance clean energy.

    5. Use predictable data-centre demand to finance new renewable generation. Condition grid connection on genuinely new (additional) renewable generation and firming, so the large, predictable demand data centres represent underwrites the clean generation the grid needs as coal retire.

  28. Appendix A: direct levy revenue and reserved compute to research, skilled-person retention, data assets and efficient Australian domain-specific models in health, agriculture, defence and public administration.

    The levy revenue and reserved capacity should be directed to the layers where durable value and control reside: research, the retention of skilled people, data assets, and the development of efficient, domain-specific models suited to Australian needs in areas such as health, agriculture, defence and public administration.

Read the original

In some quotes a program corrected obvious text-recognition errors for display.

Open in the interactive guide