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 513

Australian Council on AI Strategy (ACAIS)

Submitter group: Civil society.

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.

ACAIS argues hosting training compute is necessary for significant AI-safety influence and outsized economic benefits, but does not itself ensure safety. It proposes negotiating safety, security, research and governance conditions, biological-risk measures and energy coordination. It asks for swift resolution of the copyright impasse without favouring a legal instrument.

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 hosting future training compute is a prerequisite for significant global AI-safety influence and an outsized share of economic benefits at home.

    For Australia, which has not been a significant part of the AI supply chain, hosting future training compute is the pre-requisite for exercising any significant global influence to advance AI safety, and for capturing an outsized share of the economic benefits at home.

  2. Argues that the distinction between training and inference data centres determines whether investment buys leverage or only buildings.

    The distinction between inference and training data centres matters significantly in AI deliberations because it determines whether investment in Australia buys leverage or simply buys buildings.

  3. Asks for swift resolution of the copyright impasse, saying Australian content is used in offshore training without remuneration while current copyright laws deter training in Australia.

    We strongly urge swift resolution of the current copyright impasse because the status quo serves no one. Today, Australian content is already being used in training conducted in other jurisdictions without remuneration, while current copyright laws deter training in Australia by any AI company (large or small).

  4. Argues AI models now exploit cyber vulnerabilities at machine speed, levelling the capabilities of hackers and state actors.

    However, where once well-resourced and sophisticated operations were required to exploit cyber vulnerabilities, this is now being performed by AI models running in parallel at machine speed. Capability to act has now been levelled, with a hacker or a state actor all showing adoption of similar methodologies.

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

  5. Says biological threats receive little attention in Australian debate and warrant particular attention by the Committee.

    The risk of biological threats receives little if any attention in the Australian debate and warrants particular attention by the Committee.

  6. Says the Australian AI Safety Institute has begun testing unreleased frontier models with ASD but is small, under-resourced and reliant on goodwill.

    Commendably, the Australian AI Safety Institute, as of mid-2026, has begun testing unreleased frontier models with the Australian Signals Directorate (ASD). It is nonetheless small, under- resourced and largely dependent on the goodwill of busy staff at foreign companies and overseas safety institutes (see comparative resourcing below).

Recommendations

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

  1. Australian governments should move urgently to attract AI training on Australian terms, shaping frontier AI and sharing benefits of development and deployment.

    Australia has been dealt an extraordinary hand in attracting AI training, but it will not play itself. Australian governments must move with urgency to take advantage of this window of opportunity and give form to the distinctive contribution Australia can make to shaping frontier AI, on Australia’s terms and in a way that shares the benefits of AI development and deployment.

  2. Comprehensively examine hosting training compute across geostrategic, economic, national-security, international-governance, energy, copyright, regulatory and safety dimensions.

    Assessing the risks and benefits of hosting training compute in Australia requires comprehensive examination of what is at stake for Australia across a range of policy dimensions including, geostrategic, economic, national security, international governance, energy, copyright, regulatory and safety implications of AI.

  3. Negotiate conditions and requirements serving the national and global interest before entering a deal with frontier AI labs to host training compute.

    The Australian Government should negotiate conditions and requirements that serve our national and global interest before it enters into a deal with frontier AI labs to host training compute in Australia.

  4. Negotiate access to frontier AI models, including when compute constraints bite, as a position gained by hosting training compute.

    Strengthened national security environment: Negotiated access to frontier AI models when compute constraints bite and when powerful dual-use capabilities continue to emerge.

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

  5. Build Australian capability to assess frontier AI, including recursive self-improvement and risks of losing human control over future systems, to shape AI governance on Australian terms.

    Build AI safety knowhow: Australian assessments of frontier AI capabilities, including recursive self-improvement and the risks of losing human control over future AI systems, has become essential must-have knowledge needed to shape AI governance on Australia’s terms. No access or analytical capability like this exists in Australia.

  6. Request a specified number of AI engineers and frontier researchers under Australian jurisdiction to engage closely with government, education and industry.

    Australia can request a specified number of AI engineers and frontier researchers working under Australian jurisdiction, to engage more closely with the Australian government, education, and industry sectors.

  7. Consider requiring deals with AI companies to include AI technical staff in Australia.

    it may still be beneficial for the Australian government to ensure that deals with AI companies explicitly incorporate a requirement to have AI technical staff in Australia.

  8. Consider targeted National Innovation Visas to help AI researchers relocating to Australia move quickly, where they are not Australians living abroad.

    A research presence in Australia will be key to upskilling the local AI ecosystem and the AISI, and will generate positive spillover effects. If those researchers are not Australians living abroad, they can be supported to relocate quickly by targeted use of the National Innovation Visa.

  9. Provide Australian institutions and researchers affordable access to AI training capability for research, potentially through fellowships working directly with frontier researchers, to build analytical capability, training opportunities and direct insights.

    Some AI training capability should be available for Australian institutions and researchers to conduct AI research. This could utilise models such as Anthropic’s Claude Corps Fellowship or the Research and Safety Fellows Program where participants work directly with frontier researchers in specialised tracks. As training compute power is an increasingly valuable and scarce resource, such an arrangement would assure affordable access for Australian research purposes, giving us analytical capability, training opportunities and direct insights.

  10. Use hosting to advance AI safety science and common evaluation, incident-reporting and biological-synthesis-screening standards with democratic middle powers, avoiding a safety race to the bottom.

    A seat at the international governance table: Regardless of where AI is developed, the risks are felt worldwide. Efforts are required to jointly advance AI safety science and set global standards, rather than engage in a competitive race to the bottom, where safety is an afterthought. Having access to critical training compute opens the possibility of democratic middle powers, like Australia to convene, influence and even lead international AI governance and safety debates which can shape and pace the frontier. A seat at the table can help establish common evaluation, incident reporting and biological synthesis screening standards (see cyber and biological risks below).

  11. Explore mandatory AISI incident reporting on defined timelines as a hosting condition, covering containment failures, dangerous-capability threshold crossings and significant misuse detections, including internal models not yet public.

    A hosting arrangement could require mandatory reporting to AISI on defined timelines for incidents such as containment failures, crossing of dangerous- capability threshold crossings, and significant misuse detections, including for internal AI models not yet publicly available.

  12. Make security standards for training facilities a condition of hosting, with ASD involved in accreditation of facilities.

    Australia could make security standards for training facilities a condition of hosting, with the ASD involved in accreditation of facilities.

  13. Develop verification technologies by offering Australia as a pilot site for training-run reporting and verification regimes.

    Development of verification technologies: ideally all countries would comply with safety standard and measures prohibiting dangerous use of AI. If verification mechanisms are ever going to be tested, they need willing host jurisdictions that have training-scale facilities. Australia could offer to be the pilot site for training-run reporting and for verification regimes.

  14. Negotiate allocation of hosted compute reserved for safety and interpretability research done in partnership with AI lab researchers and external researchers.

    Australia could negotiate allocation of hosted compute to be reserved for safety and interpretability research, that is done through partnership with AI lab researchers and external researchers

  15. Attract a technical workforce to undertake research on the ground in Australia.

    Australia would need to attract a technical workforce to undertake this research on the ground in Australia.

  16. Engage in technical dialogue with China to strengthen AI safety science and measures.

    Australia’s prominent role in the Asia-Pacific also positions us well to engage in technical dialogue with China to strengthen AI safety science and measures.

  17. The Committee should give biological threats particular attention.

    The risk of biological threats receives little if any attention in the Australian debate and warrants particular attention by the Committee.

  18. Use the Director of Biosecurity’s BICON powers immediately, without new legislation, to require imported synthetic nucleic acids to come from providers screening orders for sequences of concern.

    Australia's Biosecurity Import Conditions system (BICON) gives the Director of Biosecurity the power to impose conditions on imported synthetic nucleic acids. This should be done immediately, and can be achieved without new legislation. Members of the International Gene Synthesis Consortium, already voluntarily screen orders for sequences of concern. Australia's current import permit system does not require that synthetic nucleic acids be sourced from providers that screen, and bad actors can exploit this. The current arrangement poses an unacceptable risk under the Biosecurity Act 2015. This should be remedied immediately and quickly without the need for legislative change.

  19. Build bioresilience by developing capabilities to detect novel and known pathogens early.

    Australia can also build its bioresilience by developing capabilities to detect novel and known pathogens early

  20. Pre-negotiate surge biomanufacturing capacity to accelerate production of diagnostics, therapeutics and vaccines.

    enabling swift protection by pre-negotiating surge capacity in biomanufacturing that would accelerate production of diagnostics, therapeutics and vaccines

  21. Resolve the copyright impasse urgently; ACAIS takes no position on which legal instrument Parliament should choose.

    ACAIS does not represent rightsholders or AI developers. We take no position on which legal instrument Parliament should choose in overcoming the current impasse that prevents training compute occurring in Australia by US frontier labs or any companies training AI, including Australian ones. The status quo is the one outcome on which both sides should be able to agree is a failure. Australia has neither a fair-use doctrine nor a text and data mining exception, unlike the United States, Japan and Singapore. A TDM exception was ruled out in October 2025. We strongly urge urgency in resolving the current impasse, which we understand is follows:

  22. The Committee should test the claim that frontier AI companies are license-sensitive rather than price-sensitive.

    On this account frontier AI companies are license-sensitive rather than price-sensitive. The committee should test this claim.

  23. Pursue strong, timely and productive cooperation between the Australian Government and state and territory governments to reduce 'time to power' for data centres.

    The best outcomes would also involve strong, timely and productive cooperation between the Australian Government and State and Territory governments to reduce "time to power".

  24. Set appropriate data-centre rules, improve approvals processes to keep Australia attractive for investment, and implement strong benefit-sharing arrangements when resolving copyright and securing agreement to host training at scale.

    Crafting a win-win resolution for copyright and securing in principle agreement to host training at scale in Australia will not be the final chapter in Australia’s AI journey. Setting appropriate rules for data centers, ensuring Australia remains an attractive investment destination (including improved approvals processes), and implementing strong benefit sharing arrangements will all be crucial.

Read the original

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

Open in the interactive guide