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 239

Digital Economy Council of Australia

Submitter group: Industry bodies.

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 that AI policy should remove barriers to business reorganisation, coordinate Australia's AI processes with consistent definitions, and regulate deployed systems rather than models. Recommends agent identity and attribution, verifiable disclosure pilots, selective sovereign investment and quick copyright resolution; says existing powers' sufficiency cannot yet be fully evidenced and its members differ on the copyright rule.

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 productivity gains depend on firms reorganising around AI, so rules that constrain reorganisation make red tape reform AI policy.

    AI is a general purpose technology, and general purpose technologies deliver measured productivity gains only once firms reorganise around them. Rules that fix who may supply a service, how it must be delivered and which approvals a new activity requires constrain that reorganisation. Red tape reform is therefore AI policy.

  2. Argues that the main regulatory problem is coordination rather than absent power, while stating that the sufficiency of existing powers cannot yet be fully evidenced.

    What is absent is not power but consistency,. DECA qualifies that conclusion in one respect, as we did before the Senate committee. Because Australia has no comparable dataset on AI deployment, the claim that existing powers are sufficient cannot yet be fully evidenced, and neither can the contrary claim.

  3. Argues that AI harm arises at the level of the deployed system, so regulation should target the decision pathway rather than model capability.

    Fourth, AI harm arises at deployment, at the level of the system rather than the model. Regulation directed at model capability will miss most of it; regulation directed at the decision pathway will reach it.

  4. Argues that agentic AI is the near-term governance gap and that agent identity and attribution is the tractable intervention.

    Fifth, agentic AI is the near-term governance gap. Agent identity and attribution is the tractable intervention available now, and it will be considerably harder to retrofit once agent-mediated activity is at scale.

  5. Argues that Australia should adopt global technology where it offers best value and invest selectively where it holds defensible advantage.

    Second, sovereign capability should be selective and connected end to end. Australia should adopt global technology where it offers best value and invest where it holds defensible advantage, with a pathway running from research through demonstration and procurement to export.

  6. Argues that obligations that cannot be independently verified produce documentation rather than safety.

    Sixth, obligations that cannot be independently verified produce documentation rather than safety. Verifiable disclosure is where DECA’s membership can offer the Committee something other stakeholders will not.

Recommendations

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

  1. Publish, before any AI legislation is introduced, a coordination map showing which instrument and portfolio addresses each AI issue and how Commonwealth measures interact with State and Territory frameworks.

    Publish, before any AI legislation is introduced, a coordination map identifying which AI- related issues are to be addressed by which instrument and which portfolio, and how Commonwealth measures are intended to interact with State and Territory frameworks.

  2. Treat red tape reduction as AI policy, by reinvigorating barrier-removal programs and building AI-enabled measures of regulatory complexity, discretion and approvals to set a checkable baseline of Commonwealth regulation.

    Treat red tape reduction as AI policy, by reinvigorating barrier-removal programs and building new AI-enabled measures of regulatory characteristics, including complexity, discretion and the number of approvals standing between a firm and a new activity, to establish a checkable baseline of the Commonwealth regulatory stock.

  3. Direct AI regulation at the deployed system and decision pathway, with deployer obligations proportionate to the consequence of the decision, rather than at model capability or infrastructure.

    Direct AI regulation at the deployed system and the decision pathway, applying obligations to deployers in proportion to the consequence of the decision, rather than at model capability or the infrastructure on which a model runs.

  4. Build any mandatory Australian Standards for AI on the Voluntary AI Safety Standard and AS ISO/IEC 42001, identifying the scope, threshold, commencement and binding mechanism of each obligation separately.

    Build any mandatory Australian Standards for AI on the Voluntary AI Safety Standard and AS ISO/IEC 42001, identifying the scope, threshold, commencement and binding mechanism of each obligation separately rather than adopting them as a single undifferentiated package.

  5. Address agentic AI expressly, beginning with agent identity and attribution, so a party can establish who deployed an agent, under what authority it acts and what model drives it.

    Address agentic AI expressly, beginning with agent identity and attribution, so that a party dealing with an autonomous agent can establish who deployed it, under what authority it acts and what underlying model is driving it.

  6. Design disclosure and reporting obligations for independent verification from the outset, and pilot cryptographically attested reporting of AI decision lineage with willing participants, beginning with Commonwealth AI use, before mandating a schema.

    Design disclosure and reporting obligations for independent verification from the outset, and pilot cryptographically attested reporting of AI decision lineage with willing participants, beginning with Commonwealth AI use, before mandating a schema.

  7. Concentrate sovereign AI investment in sectors of Australian comparative advantage, and connect research to commercialisation through procurement, test facilities, regulatory sandboxes and export pathways.

    Concentrate sovereign AI investment in sectors of Australian comparative advantage, and connect research to commercialisation through procurement, test facilities, well-designed regulatory sandboxes and export pathways.

  8. Resolve copyright uncertainty as a priority, through transparency on training-data governance and protection against substitutive outputs, with obligations on those who control training practices rather than downstream businesses.

    Resolve copyright uncertainty as a priority, through transparency on training-data governance and protection against substitutive outputs, with obligations falling on those who control training practices rather than on downstream Australian businesses.

  9. Publish procurement guidance and model contractual provisions for consequential AI deployments covering supplier due diligence, material post-deployment changes, audit/information rights, incident notification, liability, exit and portability.

    Publish procurement guidance and model contractual provisions for consequential AI deployments, addressing supplier due diligence, material post-deployment system changes, audit and information rights, incident notification, liability allocation, and exit and portability.

  10. Address the domestic shortage of practitioners able to govern, audit and certify AI systems, and treat assurance capability as a sovereignty question rather than only a skills question.

    Address the domestic shortage of practitioners able to govern, audit and certify AI systems, and treat assurance capability as a sovereignty question rather than only a skills question.

  11. Combine global AI adoption with domestic capability sufficient to avoid becoming solely a consumer of overseas technology.

    Australia should combine widespread adoption of leading global technologies with targeted investment in domestic capability. It does not need to develop every element of the AI supply chain, but it should retain sufficient capability, intellectual property, infrastructure and expertise to avoid becoming solely a consumer of technologies developed elsewhere.

  12. Concentrate investment in agriculture, resources, health/medical research, advanced manufacturing, financial services and defence, where expertise, datasets, demanding environments and international relationships provide advantages.

    Investment should concentrate where Australia holds deep industry expertise, valuable datasets, demanding operating environments and established international relationships. That means agriculture, resources, health and medical research, advanced manufacturing, financial services and the defence industry.

  13. Settle coordination and disclosure before identifying residual regulatory gaps against deployment data.

    The right sequence is to settle coordination and disclosure, then identify any residual gap against reported data rather than assert it in advance.

  14. Address the Privacy Act's qualified small business and employee records exemptions as a gap warranting targeted reform, since turnover does not reflect capacity to profile individuals or influence consequential decisions.

    6.2 Three gaps that warrant targeted reform First, the Privacy Act retains qualified small business and employee records exemptions. Turnover does not reflect a business’s capacity to profile individuals or influence consequential decisions, and employee information can be used well beyond conventional record administration.

  15. Introduce targeted rights to notice, an intelligible explanation and meaningful review where automated processing materially influences a consequential decision and existing arrangements provide no equivalent protection, recognising sectoral complaint mechanisms.

    DECA supports targeted rights to notice, an intelligible explanation and meaningful review where automated processing materially influences a consequential decision and existing arrangements provide no equivalent protection, recognising established sectoral complaint mechanisms.

  16. Use proportionate evidence-preservation and supplier-cooperation requirements, rather than wholesale replacement of existing liability principles, where AI makes proof of a cause of action harder.

    Third, civil liability principles require proof of the elements of a cause of action, and AI makes that evidentially harder when relevant records sit with several suppliers and the system has changed by the time the failure is investigated. Proportionate evidence-preservation and supplier- cooperation requirements are a better response than a wholesale replacement of existing liability principles.

  17. Keep nominal or rubber-stamped human oversight within regulatory scope; require accountable human decision authority calibrated to use-case risk.

    Human oversight is a factor, not an exemption. DECA defines human in the loop as a design pattern in which a clearly accountable human retains decision authority over material outcomes, through review, approval, escalation or override calibrated to the risk of the use case. Nominal or rubber-stamped oversight should not take a system out of scope.

  18. Frame AI obligations around continuous monitoring, validation and review, with re-assessment required on material change rather than on a fixed cycle alone.

    Obligations should therefore be framed around continuous monitoring, validation and review, with re-assessment required on material change rather than on a fixed cycle alone.

  19. Distinguish advisory systems from agents authorised to act; for consequential agents, address delegated authority, approval thresholds, authentication, logging, human escalation and reversing or terminating actions.

    Legal and regulatory analysis should distinguish systems that provide information or recommendations from systems authorised to take action on behalf of an organisation. Where an agent can initiate transactions, communicate externally, alter records or otherwise create legal or financial consequences, governance should address the scope of delegated authority, approval thresholds, authentication, logging, human escalation, and the ability to reverse or terminate actions.

  20. Apply foundational agent constraints prohibiting unauthorised access and requiring efficient task completion, regardless of use case.

    Working Group members have accordingly argued for limiting principles, or meta-guardrails: foundational constraints prohibiting unauthorised access and requiring efficient task completion, applied to agents regardless of use case.

  21. Adopt AI responsibly across the Commonwealth public sector to improve policy, monitoring, administration and service accessibility and demonstrate practice.

    The Commonwealth should participate in the transformation rather than acting solely as regulator, policymaker or funder. Responsible adoption of AI across the public sector could improve policy development, regulatory monitoring, program administration and the accessibility of services, help government manage rising demand within a constrained labour market, and demonstrate practice for the wider economy.

  22. Place training-governance and output-protection duties primarily on developers/providers controlling training practices rather than ordinary downstream users; protect against reproduction or material substitution of original works.

    Whatever rule is adopted should be settled quickly. DECA supports greater transparency around training-data governance and effective protection against outputs that reproduce protected expression or materially substitute for original works. Those obligations should fall primarily on model developers and providers with control over training practices, rather than on ordinary downstream business users.

  23. Enable meaningful supplier assurances without requiring business users to reconstruct training histories; avoid disproportionate costs, uncertainty and impractical tracing obligations for startups and SMEs.

    An Australian business using a third-party model should be able to obtain meaningful assurances from its supplier; it should not be expected to reconstruct that supplier’s training history. Any licensing or compliance framework should avoid imposing disproportionate costs, legal uncertainty or impractical tracing obligations on Australian startups and small and medium enterprises.

  24. Keep licensing accessible and enforcement affordable.

    Licensing must be accessible and enforcement affordable.

  25. Support standard terms and voluntary collective licensing arrangements where rights holders authorise them, covering material, permitted uses, remuneration, onward supply and termination.

    Government should support standard terms and voluntary collective arrangements where rights holders authorise them, identifying the material covered, permitted uses, remuneration, onward supply and treatment on termination.

  26. Treat moral rights separately from ownership, recognising they cannot be assigned.

    Moral rights, which are distinct from ownership and cannot be assigned, require separate treatment.

  27. Retain dataset-source and permission records proportionately, with detailed confidential evidence available when a substantiated dispute arises.

    Proportionate retention of records of dataset sources and permissions, with detailed evidence available through appropriate confidential processes when a substantiated dispute arises, is what makes a cheaper dispute forum worth having: a small claims process has limited value if a claimant still cannot obtain the information needed to establish a case.

  28. Clarify copyright protection for AI-assisted work.

    DECA also supports clarification of the protection available for AI-assisted work.

  29. Provide guidance distinguishing human contributions, including prompts, reference material, parameters and configuration, from system-generated material, with examples of evidence of human authorship businesses should retain.

    Guidance should distinguish copyright in a person’s original contributions, including prompts, reference material, parameters and configuration, from any claim over material generated by the system, with practical examples of the evidence of human authorship a business should retain.

  30. Before granting exclusive rights over wholly machine-generated material, demonstrate public benefit and consider effects on human creators and competition.

    Any proposal to confer new exclusive rights over wholly machine-generated material should first demonstrate a public benefit and consider its effect on human creators and competition.

  31. Consider AI uses expressly within the Government's work on standalone protection for First Nations traditional knowledge and traditional cultural expressions, led by First Nations peoples.

    AI uses should be expressly considered within the Government’s work on standalone protection for First Nations traditional knowledge and traditional cultural expressions, and that work should be led by First Nations peoples.

  32. Do not treat general licences as sufficient where First Nations community authority or cultural restrictions require separate consideration.

    Permission from the owner of a photograph, recording or publication may not address the cultural authority to reproduce or repurpose what it depicts or records, and general licensing arrangements should not be treated as sufficient where community authority or cultural restrictions require separate consideration.

  33. Expand Commonwealth contractual guidance to cover capability/data/rights representations, material changes, audit/information, incidents, subcontracting, data location/retention, indemnities/liability, termination, exit/portability and appropriate insurance.

    Commonwealth guidance should therefore address minimum contractual considerations for consequential AI deployments: representations about system capabilities, data use and rights, allocation of responsibility for material system changes, audit and information rights, incident notification, subcontracting, data location and retention, indemnities and limitations of liability, termination, exit and portability, and appropriate insurance.

  34. Do not make customers responsible for risks from material system changes they cannot control or meaningfully test.

    Particular attention should be given to changes made by providers after deployment. A customer should not be expected to accept responsibility for risks arising from material changes to a system that it does not control and cannot meaningfully test.

  35. Deploy AI to grow human capability; provide opt-in worker upskilling and support transitions for displaced workers.

    Consistent with DECA’s ethical principles, our position on term of reference (g) is that AI should be deployed to grow human capability rather than simply to displace it, that workers should have access to upskilling on an opt-in basis, and that productivity gains from automation carry an obligation to support transition for those displaced.

Read the original

Open in the interactive guide