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 340

Mr Peter Joseph

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.

Proposes recurring Treasury stress-testing of revenue resilience under different AI outcomes, as preparedness without a predetermined tax response. A supplementary submission suggests examining graduated independent assurance and human intervention for designated frontier compute, while comparing ownership and regulatory options, recognising limits of unilateral controls and avoiding disproportionate burdens on ordinary AI use.

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. Asks whether Australia is stress-testing the tax system's ability to raise revenue if AI shifts where taxable capacity arises.

    Is Australia systematically stress-testing whether the Commonwealth tax system will remain capable of raising the revenue future governments require if AI and automation materially alter where taxable economic capacity arises?

  2. Says the proposed test is a preparedness exercise rather than a prediction, and a failed test would trigger investigation, not automatic tax change.

    This is a preparedness exercise, not a prediction. A failed test would trigger investigation and modelling, not an automatic change in tax policy.

  3. Argues an AI-specific tax should not be the starting point, saying taxing automation merely because it replaces labour could discourage productivity-enhancing investment.

    This approach does not require an AI tax, robot tax or other tax directed specifically at technology. Taxing automation merely because it replaces human labour could discourage productivity-enhancing investment without identifying where the resulting economic capacity to bear taxation has moved.

  4. In the supplementary submission, suggests the Committee consider whether access to increasing frontier compute capability should be accompanied by increasing independent assurance of safeguards.

    I suggest the Committee consider whether, at the frontier end of that infrastructure, access to increasing capability should be accompanied by increasing independent assurance that relevant safeguards remain effective.

  5. Suggests considering whether Australia should know in advance that a technically effective means of human intervention exists for designated high-risk infrastructure, and who can authorise it.

    The narrower question is whether Australia should know in advance that a technically effective means of human intervention exists for designated high-risk infrastructure, who can authorise its use, and under what circumstances.

  6. Says that Australian regulation of compute cannot manage global AI risk alone and cannot substitute for international cooperation.

    First, Australia cannot manage global AI risk simply by regulating Australian compute. Models, expertise and computational resources can be located overseas. Australian measures therefore cannot substitute for international cooperation.

Recommendations

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

  1. Make failed revenue-resilience tests trigger investigation and modelling rather than automatic tax changes; permit findings that no material change is needed.

    This is a preparedness exercise, not a prediction. A failed test would trigger investigation and modelling, not an automatic change in tax policy. If the existing tax mix continued to perform adequately across plausible scenarios, the test could equally conclude that no material change was required.

  2. Have Treasury establish recurring revenue-resilience scenario tests across AI/automation outcomes, covering employment, labour income, productivity, profits, capital returns, consumption, economic rents and taxable-capacity shifts, with change indicators and further modelling/research needs.

    The Committee should recommend that the Australian Treasury establish a recurring Future Revenue Resilience Test that uses explicit scenarios to assess whether the Commonwealth tax system remains capable of raising required revenue under materially different AI and automation outcomes. The analysis should cover employment, labour income, productivity, enterprise profits, capital returns, consumption, economic rents and changes in where taxable capacity arises. It should also identify observable indicators of structural change and areas requiring further modelling or policy research.

  3. Consider revenue-resilience scenarios in Intergenerational Reports, associated Treasury analysis, standalone publications or other periodic fiscal processes; recurring institutional responsibility matters more than publication choice.

    Australia already has an institutional mechanism for considering long-term structural economic and fiscal change through the Intergenerational Report. Future reports, or associated Treasury analysis, could include explicit revenue-resilience scenarios. This is a placement option rather than a design requirement: the test could also sit in a standalone Treasury publication or another periodic fiscal process. The essential feature is recurring institutional responsibility, not the choice of publication.

  4. Consider independently tested, effective safeguards before materially greater frontier capability is enabled, potentially covering model evaluation, cybersecurity, access controls, incident reporting, containment and human intervention.

    Before materially greater capability is enabled, an operator might be required to provide evidence that safeguards appropriate to that capability have been independently tested and remain effective. Depending on the circumstances, that could include evidence concerning model evaluation, cybersecurity, access controls, incident reporting, containment arrangements and the ability for humans to intervene.

  5. Avoid disproportionate assurance obligations on ordinary computing, smaller developers and research.

    This approach would also need to avoid imposing disproportionate obligations on ordinary business computing, smaller developers and research.

  6. Periodically review any computational thresholds as hardware and algorithms improve.

    Second, compute efficiency will improve. A fixed computational threshold that identifies frontier capability today may become inappropriate as hardware and algorithms improve. Any threshold-based regime would therefore require periodic review.

  7. Distinguish genuinely consequential frontier capability from ordinary AI development/use to avoid assurance costs entrenching incumbents.

    Finally, assurance requirements themselves can entrench large incumbents if compliance costs make it difficult for smaller competitors to participate. Any regime should therefore distinguish genuinely consequential frontier capability from ordinary AI development and use.

  8. Consider frontier-scale compute within the approach to sovereign AI capability and strategic resilience.

    whether frontier-scale compute should be expressly considered within Australia's approach to sovereign AI capability and strategic resilience

  9. Consider whether designated AI capability levels should attract progressively stronger independent assurance.

    whether designated levels of AI capability should attract progressively stronger independent assurance requirements

  10. Consider whether publicly funded, supported or otherwise designated strategic AI infrastructure could incorporate Preventative Assurance requirements before materially greater capability is enabled.

    whether publicly funded, supported or otherwise designated strategic AI infrastructure could incorporate Preventative Assurance requirements before materially greater capability is enabled

  11. Consider requiring designated frontier operators to demonstrate effective human-intervention and workload-isolation mechanisms, with clearly defined legal authority for exceptional use.

    whether operators of designated frontier infrastructure should be required to demonstrate effective human intervention and workload-isolation mechanisms, together with clearly defined legal authority governing their exceptional use

  12. Consider the relative advantages and disadvantages of public ownership, co-investment, licensing, contractual conditions and conventional regulation as mechanisms for these objectives.

    the relative advantages and disadvantages of public ownership, co-investment, licensing, contractual conditions and conventional regulation as mechanisms for achieving these objectives

  13. Consider how an Australian framework could operate alongside international approaches, recognising the limits of unilateral controls.

    how an Australian framework could operate alongside international approaches, recognising the limits of unilateral controls

Read the original

Open in the interactive guide