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 280

Australian Strategic Policy Institute (ASPI)

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.

Authors at ASPI, stating their views do not reflect a single ASPI perspective, argue Australia should pursue AI benefits through data centre investment and local capability while building public trust, strengthening security and shaping global safety governance. They recommend frontier-safeguard diplomacy, a better-funded AI Safety Institute, rigorous data-centre standards, talent and sovereign compute investment, and updated critical infrastructure cyber standards.

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. Says the government's plan to attract AI computing investment and build a higher-value industry at the model and application layers is a sound strategy, with the challenge now in execution.

    The federal government’s proposal to do so by attracting AI computing infrastructure investment and using that foundation to build a higher value industry at the model and application layers is a sound strategy. The challenge now is in the execution

  2. Describes AI sovereignty as keeping agency and strategic options by controlling enough core inputs not to be denied or coerced.

    The end goal is to strengthen Australian AI sovereignty—a much discussed term that has different interpretations but boils down to ensuring we keep our agency and strategic options by controlling enough of the core inputs that we cannot be denied or coerced.

  3. Argues that risks from superintelligence or powerful general intelligence are very real and must not be dismissed.

    The risks posed by superintelligence or powerful general intelligence are very real and must not be dismissed. The stunning pace of AI development, with growth both in the size and power of models, and in the autonomy and independence of agents, is outstripping researchers’ ability to ensure the technology is safe.

  4. Argues that hosting data centres that train frontier models should give Australia leverage to negotiate access to those models.

    building data centres that can be used to train new frontier models should give Australia leverage to negotiate access to those models. This access is not otherwise assured, even from Australia’s ally the United States.

  5. States it is neutral on the copyright mechanism provided investment is not deterred, but says a compromise involving updates to the regime is warranted.

    We are neutral on the precise mechanism to resolve the copyright issue provided it does not deter investment. Our only overarching observation is that a compromise involving updates to the regime is warranted

  6. Says AI is lowering skill barriers to sophisticated cyber attacks, citing experts' warnings that nation-state capabilities are within reach of individuals with modest skills.

    AI is dramatically lowering the skill barriers to sophisticated cyber attacks, with some experts warning that capabilities previously confined to nation-states are now within reach of individuals with only modest skills.

Recommendations

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

  1. Use Australian diplomacy to push the US and China towards mandatory frontier-AI safeguards, including safeguards calibrated to increasing model capability ('pacing the frontier').

    Use Australian diplomacy to push the US and China towards mandatory frontier-AI safeguards, including safeguards calibrated to increasing model capability (“pacing the frontier”).

  2. Increase funding for the AI Safety Institute so it can recruit leading experts and contribute to global safety and alignment research, with capabilities including independent frontier model evaluation, cyber/bio capability assessment, incident analysis, international information sharing and pre-deployment model access.

    Increase funding for Australia’s AI Safety Institute, enabling it to recruit leading experts and contribute more substantially to global safety and alignment research. Particular capabilities should include independent frontier model evaluation; cyber/bio capability assessment; incident analysis; international information sharing; and access to models before deployment.

  3. Work with other democracies to lead AI capability so that democratic principles of technology governance diffuse globally.

    Work with other democracies to lead AI capability and therefore democratic principles of technology governance diffuse globally.

  4. Implement the government's data-centre standards rigorously, especially requirements to support Australian research, supply affordable compute, build local skills and invest in local supply chains.

    Implement the government’s data-centre standards rigorously, especially requirements for providers to support Australian research, supply affordable compute, build local skills and invest in Australian supply chains.

  5. Bolster programs to attract, develop and retain AI talent.

    Bolster programs to attract, develop and retain AI talent, treating skills and expertise as a particularly durable investment given uncertainty about AI’s development.

  6. Increase federal public investment in AI, including consideration of dedicated sovereign compute and model-development capability.

    Increase federal public investment in AI, including consideration of dedicated sovereign compute and model-development capability.

  7. Use Australian data-centre investment as leverage to secure access to advanced US frontier models, particularly for national security, critical infrastructure and strategic industries.

    Use Australian data-centre investment as leverage to secure access to advanced US frontier models, particularly for national security, critical infrastructure and strategically important industries.

  8. Fund Australian research into defences against AI-enabled biological threats.

    Fund Australian research into defences against AI-enabled biological threats, drawing on Australia’s strengths in medical and biological sciences.

  9. Keep mandatory cyber-security standards, especially for critical infrastructure, continually updated as AI capabilities evolve, including through the SOCI Act framework.

    Keep mandatory cyber-security standards—especially for critical infrastructure—continually updated as AI capabilities evolve, including through the SOCI Act framework.

  10. Explore in detail with the government and industry how the national data-centre standards will reliably translate into the desired national interest outcomes.

    We recommend the committee explore in detail with the government and industry how the government will ensure the national standards translate reliably into the desired national interest outcomes.

  11. Grow Australia's model development capacity at a realistic level, including by fine-tuning and enhancing off-the-shelf open weight models for bespoke purposes, while noting that many of the best open weight models are made by Chinese companies, which it says pose security and strategic risks and should be considered too risky for national security and critical infrastructure uses.

    But Australia can and should grow its model development capacity at a more modest and realistic level, including by taking off-the-shelf open weight models and fine-tuning and enhancing them for our own bespoke purposes. One note of caution on this ambition is that many of the best open weight models currently are made by Chinese companies, which pose security and strategic risks and would need to be considered very carefully by Australian users. Certainly for national security and critical infrastructure uses, they should be considered too risky.

  12. Reach a prompt, reasonable copyright compromise that does not deter investment, within the finite opportunity for Australian hosting of US model training.

    We urge all parties to reach a reasonable compromise as soon as possible given that the window of opportunity for Australia to become the natural second home for US companies to train models is finite. We are neutral on the precise mechanism to resolve the copyright issue provided it does not deter investment.

  13. Highlight the Australian Signals Directorate's work so the private sector and public are aware of the resources available.

    The Australian Signals Directorate is doing a superb job of managing the cyber threat transformation and we encourage the Committee to highlight the Directorate’s work to ensure the private sector and public are aware of the resources available.

  14. Continue monitoring the appropriate level of resourcing for agencies such as the ASD so they keep pace with AI.

    the government should continue to monitor the appropriate level of resourcing for agencies such as the ASD to ensure they can keep pace with the progress of the technology and the scale of its impacts.

  15. Be wary of arguments for rapid frontier AI release and use diplomatic channels to urge caution wherever possible.

    Australia should therefore be very wary of arguments in favour of rapid frontier AI release and should use diplomatic channels to urge caution wherever possible.

  16. Carefully consider extraterritorial liability laws holding overseas AI companies accountable when agents attack Australia without company direction or knowledge, if behaviour is deemed reckless.

    One consideration for the government and Parliament is to review legal liability laws so that an overseas-based AI company whose agents carried out a cyber attack in Australia without the company’s direction or knowledge could be held accountable if their behaviour were deemed reckless. Given such laws would need to apply extra-territorially, this is a tricky area that would need to be considered carefully, noting that even under current US law, OpenAI doesn’t appear to have committed any crime and Hugging Face has no stated plans to pursue damages or legal accountability.

  17. Continue working with states and territories to keep laws up to date to protect Australians, especially children and other vulnerable groups, from AI harms; criminalising deepfake pornography and CSAM is given as an example.

    the Commonwealth should continue working closely with states and territories to keep laws up to date to protect Australians—especially children and other vulnerable groups—from more immediate harms created by AI. Efforts to criminalise behaviours such as the creation and distribution of deepfake pornography and CSAM material are clear examples.

  18. Continue implementing the independent SOCI Act review findings and update them regularly as needed.

    We therefore support fully the independent review into the Security of Critical Infrastructure Act 2018 and the ongoing implementation of its findings. And we encourage further regular updates as needed, given the speed of progress in AI.

  19. Advocate to the US and Chinese governments and more widely to other countries for greater global cooperation on AI safety.

    The dismissive attitude shown by both the US and Chinese governments about serious safety concerns being expressed by the US industry and many independent experts is a problem over which the Australian government has limited influence, but it should nonetheless advocate to those governments and more widely to other countries for greater global cooperation on safety.

  20. Continue, and if necessary strengthen, the government's emphasis on building social licence.

    The government’s commendable emphasis on building social licence should continue and, if necessary, be strengthened.

  21. Bolster efforts to address all legitimate concerns about AI, including by regulating where that can be effective.

    Therefore the Australian government should bolster its efforts to address all legitimate concerns, including by regulating where that can be effective.

  22. Continue Coalition support for data-centre construction.

    And the Coalition’s equally commendable support for data centre construction should likewise continue.

Read the original

Open in the interactive guide