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 17

Shop, Distributive and Allied Employees' Association (SDA)

Submitter group: Unions.

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, drawing on member experience, survey research and two commissioned reports, that workplace AI should be governed through work health and safety duties, consultation and Fair Work institutions, and that Australia should build sovereign compute and model assurance capability. It states it does not oppose technology adoption in principle, and makes sixteen numbered recommendations including mandatory guardrails.

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. Describes algorithmic rostering, automated task allocation based on assumptions rather than actual work, and workplace change introduced with little consultation.

    algorithmic rostering that removes control over their working hours, automated task allocation systems built on assumptions about how work is done rather than the reality of doing it, and workplace change introduced with little or no consultation before decisions are finalised.

  2. Says its 11,000-plus-worker retail survey found 72% reported burnout and 77% faced high psychosocial-harm risk, with inadequate consultation a major contributing hazard.

    Among the more than 11,000 retail workers surveyed for the SDA's 2025 Under Pressure report, 72 per cent reported experiencing burnout and 77 per cent were assessed at high risk of psychosocial harm, with inadequate consultation on workplace change identified as one of the most significant and persistent contributing hazards.

  3. Argues, drawing on a commissioned report, that sovereignty depends on whether Australia can design, host, govern and secure essential AI systems, not only on where data centres are located.

    Sovereignty in the AI era is therefore determined not by where data centres sit, but by whether a nation can design, host, govern and secure the systems that underpin its essential functions.

  4. Says, citing a report, that frontier models and cloud platforms are concentrated in a few foreign firms and access to essential AI capability can be revoked from outside Australia.

    Frontier models, cloud platforms and data pipelines are concentrated in a handful of foreign firms subject to foreign political authority. As the report shows, access to essential AI capability can be revoked, restricted or repriced by decisions made outside Australia.

  5. Argues that voluntary compliance regimes have under-delivered in comparable digital transitions, so workers should not bear a second retrofitted-regulation cycle.

    Voluntary compliance regimes have consistently under-delivered in comparable digital transitions, and workers should not bear the cost of a second retrofitted-regulation cycle of the kind seen with social media.

  6. Says WHS duties require eliminating health and safety risks or, where not reasonably practicable, minimising them, including excessive workloads, unsafe staffing and poorly managed change.

    Under the Work Health and Safety Act, employers hold a primary duty to eliminate or, where that is not reasonably practicable, minimise risks to workers' health and safety, and this duty explicitly extends to psychosocial hazards such as excessive workloads, unsafe staffing levels, and poorly managed change.

Recommendations

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

  1. Recognise staffing, rostering, task allocation, productivity monitoring, recruitment, progression, promotion, discipline or termination decisions generated or materially informed by AI as engaging the WHS psychosocial-hazard duty.

    Where staffing, rostering, task allocation, productivity monitoring, recruitment, progression, promotion, discipline or termination decisions are generated or materially informed by AI systems, this should be explicitly recognised as engaging the WHS Act's primary duty to eliminate or minimise psychosocial hazards.

  2. Stop industrial tribunals treating staffing levels as unreviewable managerial discretion where evidence shows understaffing is a foreseeable and identified psychosocial hazard.

    Staffing levels should no longer be treated by industrial tribunals as a matter of unreviewable managerial discretion where the evidence shows understaffing is a foreseeable and identified psychosocial hazard.

  3. Require realistic, repeatable, long-term sustainable AI work measurements, excluding short peak-period benchmarks; account for required breaks and prohibit discriminatory work allocation or decisions.

    To give this duty practical effect, legislators and regulators should require that AI driven systems used for these purposes be built on realistic, repeatable and long-term sustainable measurements of work (not benchmarks derived from short, peak-period time-and-motion studies), that they factor in required break times, and that they not allocate work or make decisions based on discriminatory attributes.

  4. Give WHS entry permit holders appropriate rights to access and inspect relevant digital work systems when investigating suspected contraventions.

    WHS entry permit holders should have appropriate rights to access and inspect relevant digital work systems when investigating suspected contraventions.

  5. Have SafeWork Australia and state and territory regulators update codes of practice to reflect these requirements specifically for AI and algorithmic systems.

    SafeWork Australia and state and territory regulators should update relevant codes of practice to reflect these requirements specifically for AI and algorithmic systems.

  6. Require industrial agreements to include advance notice of AI-driven change, rights to explanation of automated employment decisions, and guaranteed retraining and redeployment pathways, with redundancy a last resort.

    Industrial agreements should be required to include advance notice of AI-driven organisational change, rights to explanation for automated decisions affecting employment, and guaranteed access to retraining and redeployment pathways, with redundancy treated as a last resort rather than a default response to automation.

  7. Have SafeWork Australia and state and territory WHS regulators collect and publicly report data on workers' compensation claims, burnout and psychosocial hazards attributable to AI-enabled workplace systems.

    SafeWork Australia and state and territory WHS regulators should collect and publicly report data on workers' compensation claims, burnout and psychosocial hazard exposure specifically attributable to AI-enabled workplace systems

  8. Examine the tripartite Retail Employee Safety Council model as a template for extending this monitoring function to other AI-exposed sectors.

    SafeWork Australia and state and territory WHS regulators should collect and publicly report data on workers' compensation claims, burnout and psychosocial hazard exposure specifically attributable to AI-enabled workplace systems, so that future policy responses are grounded in sector wide evidence. The tripartite Retail Employee Safety Council model should be examined as a template for extending this monitoring function to other AI-exposed sectors.

  9. Establish a specialist Expert Panel on AI within the Fair Work Commission to set standards for and oversee workplace AI, including rostering, surveillance, performance management, hiring and termination.

    A specialist Expert Panel on AI should be established within the Fair Work Commission, tasked with setting standards and overseeing the use of AI in workplace relations, including rostering, surveillance, performance management, hiring and termination.

  10. Require statutory AI Implementation Agreements with workers and representatives before high-impact deployment, covering consultation, job security, retraining, transparency, privacy and data use; prohibit employer discretion or bargaining away this prerequisite.

    Employers should be required to establish statutory AI Implementation Agreements with workers and their representatives before deploying high impact AI systems, covering consultation, job security, retraining, transparency, and privacy and data use, operating as a precondition to deployment rather than a matter left to employer discretion or bargained away through a general enterprise agreement.

  11. Underpin the AI Implementation Agreement framework with national AI legislation supported by a National AI Authority.

    Employers should be required to establish statutory AI Implementation Agreements with workers and their representatives before deploying high impact AI systems, covering consultation, job security, retraining, transparency, and privacy and data use, operating as a precondition to deployment rather than a matter left to employer discretion or bargained away through a general enterprise agreement. This framework should be underpinned by national AI legislation and supported by a National AI Authority,

  12. Apply WHS consultation, so far as reasonably practicable before decisions, to AI introduction or material alteration: assess risks with workers, safety representatives and unions at every workplace, rather than chain-wide pilot sign-off.

    The WHS Act's duty to consult, so far as reasonably practicable, before workplace decisions are made should be applied explicitly to the introduction or material alteration of AI and algorithmic systems, with risk assessments and genuine engagement of workers, health and safety representatives and unions conducted at each individual workplace rather than a single pilot site whose sign-off is then applied chain-wide.

  13. Require employers to engage workers and representatives during design and testing, before implementing AI materially affecting task allocation, monitoring or performance evaluation, so systems reflect actual work.

    Employers deploying AI systems that materially affect task allocation, monitoring or performance evaluation should be required to engage workers and their representatives at the design and testing stage, not only at implementation, so that systems reflect the reality of how work is actually done rather than assumptions about how it is imagined to be done

  14. Update privacy, surveillance and workplace technology protections to give workers enforceable rights to know when AI is used in decisions affecting them, receive an explanation of automated decisions affecting their employment, and challenge those decisions.

    Privacy, surveillance and workplace technology protections should be updated to give workers enforceable rights to know when AI is used in decisions affecting them, to receive an explanation of automated decisions affecting their employment, and to challenge those decisions

  15. Where automated systems contribute to discipline or performance outcomes, disclose their standards to workers in advance and require meaningful human assessment of individual circumstances before disciplinary consequences.

    Where an automated system contributes to a disciplinary or performance outcome, the standard it applies must be disclosed to the worker in advance, and a human decision-maker must meaningfully assess individual circumstances before any disciplinary consequence follows.

  16. Before biometric surveillance deployment, require worker and union consultation, rapid deletion of non-matching data, and independently verifiable retention, destruction and audit requirements.

    Deployment of biometric surveillance technology, including facial recognition, must be preceded by genuine consultation with workers and their union, use rapid-matching systems that delete non-matching data almost instantly, and be subject to independently verifiable retention, destruction and audit requirements.

  17. Invest in sovereign compute and cloud for defence and essential services, insulated from foreign jurisdictional claims; classify never-offshore, preferred-hosting and standard data, with appropriate AUKUS and Five Eyes carve-outs.

    The Commonwealth should invest in sovereign compute and cloud infrastructure for defence and essential services, insulated from foreign jurisdictional claims, and adopt a tiered data sovereignty classification system distinguishing data that must never leave Australian-governed infrastructure from data subject to preferred or standard hosting arrangements, with appropriate carve-outs for AUKUS and Five Eyes obligations.

  18. Establish a sovereign, standing technical body for adversarial testing, continuous verification and training-data auditing of AI used in defence and critical infrastructure.

    A sovereign, standing technical body should be established to conduct adversarial testing, continuous verification of model behaviour, and training-data auditing for AI systems used in defence and critical infrastructure, so that Australia is not required to trust foreign-governed systems it cannot independently verify.

  19. Government, industry and unions to establish a National AI Sovereignty Compact, chaired by the Department of the Prime Minister and Cabinet, with binding rather than advisory authority.

    Government, industry and unions should establish a National AI Sovereignty Compact, chaired by the Department of the Prime Minister and Cabinet, with binding rather than advisory authority.

  20. Have the Compact align departmental compute, assurance, workforce and R&D investment; set standards for defence, essential services and critical democratic systems; condition Commonwealth procurement eligibility on sovereign capability and labour standards for critical-infrastructure AI.

    The Compact should align sovereign compute, model assurance, workforce and R&D investment across departments, set sovereign capability standards for defence, essential services and critical democratic systems, and govern procurement conditions so that firms deploying AI in critical infrastructure must demonstrate compliance with sovereign capability and labour standards as a condition of Commonwealth eligibility.

  21. Government, unions, research institutions and industry to establish a National AI Skills Academy to design and deliver AI and cybersecurity training pipelines, embedding organised labour in curriculum design.

    Government, unions, research institutions and industry should establish a National AI Skills Academy to design and deliver AI and cybersecurity training pipelines, embedding organised labour in curriculum design so that training reflects the conditions in which workers actually operate AI systems.

  22. Have the Academy scale FSO's Skills Accelerator-AI and training-package reforms rather than duplicate them, under tripartite governance giving government, industry and unions equal standing.

    This Academy should build on and scale the work already underway through the Future Skills Organisation, including its Skills Accelerator-AI and existing training package reforms recognised in the Commonwealth's National AI Plan, rather than duplicating this infrastructure, with formal tripartite governance giving government, industry and unions equal standing over its future direction.

  23. Include portable, stackable micro-credentials accessible to casual and part-time workers and to workers in regional Australia.

    This Academy should build on and scale the work already underway through the Future Skills Organisation, including its Skills Accelerator-AI and existing training package reforms recognised in the Commonwealth's National AI Plan, rather than duplicating this infrastructure, with formal tripartite governance giving government, industry and unions equal standing over its future direction. This should include portable, stackable micro-credentials accessible to casual and part-time workers and to workers in regional Australia.

  24. Make the ten voluntary guardrails in the Guidance for AI Adoption mandatory for AI systems that materially affect employment, safety or consumer welfare.

    The ten voluntary guardrails set out in the Commonwealth's Guidance for AI Adoption should be made mandatory for AI systems that materially affect employment, safety or consumer welfare

  25. Legislate that responsibility for AI-generated content, including content published without human moderation, rests with the platform, developer or deployer rather than the technology or an end user without control over publication.

    Legislation should make clear that responsibility for AI generated content, including content automatically published without human moderation, rests with the platform, developer or deployer, not with the technology itself or with an end user who did not control publication.

  26. Fund regionally distributed AI and automation hubs linked to TAFEs, universities and industry clusters, with targeted support for small and family businesses facing cost, skills and infrastructure barriers.

    Government should fund regionally distributed AI and automation hubs linked to TAFEs, universities and industry clusters, together with targeted support to help small and family businesses overcome the cost, skills and infrastructure barriers that currently concentrate AI adoption among large firms and metropolitan centres.

  27. Target NRF, Defence Industry CRC, CSIRO and university R&D funding at Australian-data sovereign AI in sectoral-advantage areas including energy, water and defence, with grant weighting retaining Australian-owned capability.

    Public R&D investment, through the National Reconstruction Fund, the Defence Industry Cooperative Research Centres program, CSIRO and university partnerships, should target domain specific sovereign AI systems trained on Australian data in areas of sectoral advantage such as energy, water and defence, with sovereign capability weighting embedded in grant conditions so that publicly funded research generates Australian-owned capability rather than subsidising foreign platforms.

  28. Properly resource the Australian AI Safety Institute, sector regulators and the Fair Work Commission to independently assess model assurance claims and identify emerging risks.

    The Australian AI Safety Institute, sector regulators and the Fair Work Commission should be properly resourced to independently assess model assurance claims and identify emerging risks, rather than relying on the assurances of the firms whose systems they are assessing.

  29. Require independently verifiable oversight of staffing outcomes rather than employer self-assessment.

    which is precisely why independently verifiable oversight of staffing outcomes, not employer self-assessment, is necessary to give a WHS duty of this kind any real effect.

  30. Ensure consultation covers procurement, design, piloting, deployment and failure modes, including continuity during system downtime.

    A consultation process genuinely covering procurement, design, piloting and deployment would need to account for failure modes as well as normal operation, not treat system downtime as someone else's problem to fix.

  31. Allow retail facial recognition only to protect workers and customers from aggression and violence, under enforceable freedom-of-association, informed-consent, data-use and privacy safeguards.

    The SDA supports the use of facial recognition technology in retail stores, but only for the purpose of protecting workers and customers from aggressive and violent incidents, and only where its use is governed by robust, enforceable guardrails protecting the right to freedom of association, informed consent to its use, restrictions on the use of data and the privacy of both workers and customers.

  32. Require regular independent audits of how matched and retained biometric data is used.

    Protecting that social licence requires enforceable guardrails specific to biometric data, such as rapid-matching systems that delete non-matching data almost instantly rather than retaining it, regular independent auditing of how matched and retained data is used, and, as with any significant workplace technology change, genuine consultation with workers and their union before deployment, not after a regulator has already had to intervene.

  33. Embed accessible AI learning in TAFEs, local and community providers, without requiring workers to relocate or leave employment.

    Universal access to training is essential. Workers in regional Australia, outer-suburban logistics hubs and smaller retail centres must be able to acquire AI literacy and digital capability through accessible, portable and stackable learning pathways. This includes embedding AI skills in TAFEs, local training providers and community-based institutions so that workers do not have to relocate or leave employment to participate.

  34. Embed Australian organisations at the model-development layer in AUKUS Pillar II AI co-development, rather than only as downstream integrators of finished allied systems.

    The partnership's Pillar II advanced capabilities stream offers a structured pathway to co-develop sovereign defence AI systems across Intelligence, Surveillance, and Reconnaissance (ISR) analytics, autonomous platform navigation, logistics optimisation and cyber defence automation, provided co-development embeds Australian organisations at the model development layer rather than as downstream integrators of finished allied systems.

  35. Use procurement frameworks prioritising Australian-owned innovation and domestic intellectual property from public investment.

    Sovereign capability must be treated as a strategic national asset, with procurement frameworks that prioritise Australian-owned innovation and ensure that public investment generates domestic intellectual property rather than subsidising foreign platforms.

  36. Provide a Universal AI Skills Guarantee: public AI/digital-literacy training, portable pro-rata casual and part-time training rights, paid employer-directed training time, and dedicated support for sole traders and small business.

    A Universal AI Skills Guarantee, ensuring publicly supported AI and digital literacy training, portable pro-rata training rights for part-time and casual workers, paid time for employer-directed training, and dedicated support for sole traders and small business, would deliver this directly.

  37. Deliver dedicated support for sole traders and small businesses via the states under the Universal AI Skills Guarantee.

    A Universal AI Skills Guarantee, ensuring publicly supported AI and digital literacy training, including portable training rights for part-time and casual workers and dedicated support via the states for sole traders and small business, would directly address this gap.

  38. Create a government-co-invested Sovereign Cloud and Compute Program coordinated with relevant agencies, prioritising Australian operational control, security assurance and long-term supply resilience for strategic workloads.

    A dedicated Sovereign Cloud and Compute Program, co-invested by government in coordination with relevant agencies, prioritising Australian operational control, security assurance and long-term supply resilience for strategic workloads, would provide the foundation for local innovation

  39. Use targeted skilled migration for cleared AI engineers, systems integrators and model-assurance specialists, partnerships embedding Australian workers in sovereign-capable domestic firms, and streamlined security clearances.

    In the near term, this will likely require targeted skilled migration pathways for cleared AI engineers, systems integrators and model assurance specialists, industry partnership programs that embed Australian workers in sovereign capable domestic firms and streamlined security clearance pipelines.

  40. Establish a national sovereign AI research network coordinating public agencies, university AI institutes and relevant government science and technology bodies to develop research and trained sovereign practitioners.

    Over the longer term, a national sovereign AI research network coordinating investment across public research agencies, university AI institutes and relevant government science and technology bodies would create the institutional spine for sustained R&D, one designed to produce not just research output but trained sovereign practitioners.

  41. Retain Australian ownership of Commonwealth-funded AI model weights, training data and architectural innovations; when international partnerships are necessary, preserve rights to audit, adapt and redeploy outputs without foreign permission.

    Commonwealth funded AI research should retain ownership of model weights, training data and architectural innovations within Australia, and where partnerships with international firms are necessary, agreements must preserve Australia's right to audit, adapt and redeploy model outputs without seeking foreign permission.

  42. Expand the National Reconstruction Fund and Defence Industry CRC instruments for AI capability development, with sovereign-capability weighting in grants.

    existing instruments such as the National Reconstruction Fund and the Defence Industry Cooperative Research Centres program offer channels for long-term capital that should be expanded to fund AI capability development, with sovereign capability weighting built into grant criteria.

  43. Tie sovereign AI funding to social aims and worker-focused standards, protecting job quality and workplace consultation.

    This public investment should not be treated as an industrial question alone: consistent with the government's own approach through Future Made in Australia, sovereign AI funding should be tied to social aims, ensuring new technologies are held to the same worker-focused standards as established industries rather than allowing capability gains to come at the cost of job quality or workplace consultation.

  44. Classify defence, critical-infrastructure and essential-service data for Australian-governed infrastructure and prohibit it leaving Australian jurisdiction.

    Data supporting defence, critical infrastructure or essential services should be classified as requiring Australian governed infrastructure and must never leave Australian jurisdiction under any circumstances.

  45. For sensitive lower-risk data, prefer Australian hosting and require Australian government approval for limited offshore processing under strict conditions; retain existing arrangements for standard commercial data.

    A second tier of sensitive but lower risk data should be subject to preferred hosting arrangements, requiring Australian government approval but allowing limited offshore processing under strict conditions. A third tier of standard commercial data can continue under existing arrangements.

  46. Include appropriate AUKUS and Five Eyes data-sharing carve-outs in the classification system to honour alliance interoperability obligations.

    This classification system should include appropriate carve outs for data sharing obligations under AUKUS and Five Eyes arrangements, since Australia's alliance commitments require some degree of interoperability with partner intelligence and defence systems that a blanket restriction on offshore data would make impossible to honour.

  47. Have the Skills Academy coordinate training through TAFEs, unions, community providers and industry partnerships reflecting workplace conditions, building on FSO infrastructure.

    Skills development must be embedded in institutions that workers trust, including TAFEs, unions, community-based providers and industry partnerships that reflect real workplace conditions, with the Academy coordinating this network, building on FSO's existing infrastructure, rather than duplicating it.

  48. Require AI Implementation Agreements covering high-impact hiring, rostering, performance, surveillance, discipline or termination systems, with early notification, design input, data-input and decision-parameter information, and meaningful human review.

    A requirement for employers to establish AI Implementation Agreements with workers and their representatives before deploying high-impact AI systems, covering hiring, rostering, performance management, surveillance, discipline or termination, would formalise this, guaranteeing early notification, worker input into system design, clear information on data inputs and decision-making parameters, and meaningful human review of significant automated decisions.

  49. Make AI Implementation Agreements a statutory pre-deployment requirement alongside enterprise agreements; strengthen existing AI terms without duplication or undercutting, and require consultation and agreement regardless of an enterprise agreement's existence or content.

    Where an enterprise agreement already contains AI specific terms, the AI Implementation Agreement process should build on and strengthen those terms rather than duplicate or undercut them, but the statutory requirement to consult and agree before deployment should apply regardless of whether an enterprise agreement exists or what it says.

  50. Invest long-term in education, apprenticeships, research and mid-career pathways, alongside targeted migration and clearance-ready training to address near-term shortages of cleared AI engineers and model-assurance specialists.

    This requires long-term investment in education, apprenticeships, research pathways and mid-career transitions that allow workers to move into high-value roles as industries evolve, alongside targeted skilled migration and clearance ready training pipelines to address acute near-term shortages of cleared AI engineers and model assurance specialists.

  51. Establish a government-led AI in Industrial Relations Taskforce with unions, employers and sole traders; review Fair Work Act coverage of AI, algorithmic management and automated decisions, with terms preventing employer-led deregulation.

    A dedicated AI in Industrial Relations Taskforce, jointly led by government with formal participation from unions, employer representatives and sole-trader bodies, should oversee this transition, including a comprehensive review of the Fair Work Act to address the risks posed by AI, algorithmic management and automated decision-making, with terms of reference designed to prevent the process becoming a vehicle for employer-led deregulation.

  52. Have this Committee's findings draw on the Public Accounts and Audit Committee's recommendations for government working groups, consistent AI governance and clearer accountability and risk mitigation.

    This is not a new concern for this Parliament: the Joint Committee of Public Accounts and Audit's Proceed with Caution inquiry into AI governance across Commonwealth departments recommended whole-of-government working groups, consistent governance frameworks and clearer accountability and risk-mitigation mechanisms as AI systems are deployed in public service operations, recommendations that should inform this Committee's own findings.

  53. Require government AI transparency to include plain-language explanations at the point of decision and meaningful review avenues.

    This is not a new concern for this Parliament: the Joint Committee of Public Accounts and Audit's Proceed with Caution inquiry into AI governance across Commonwealth departments recommended whole-of-government working groups, consistent governance frameworks and clearer accountability and risk-mitigation mechanisms as AI systems are deployed in public service operations, recommendations that should inform this Committee's own findings. Opaque or unexplainable systems erode procedural fairness and can produce errors that are difficult to detect or challenge. Transparency must therefore extend beyond technical documentation to include mandatory plain-language explainability at the point of decision, and meaningful avenues for review.

  54. Keep deploying government agencies accountable, require substantive human powers to interrogate or override AI, independent performance/bias/outcome audits, and enforceable appeal and redress.

    Government agencies must remain responsible for the outcomes of the AI systems they deploy, including those procured from external vendors. Automated decisions cannot be treated as neutral outputs of technology; they are administrative actions with legal and social consequences. Human-in-the-loop mechanisms must function as substantive safeguards rather than formalities, ensuring public servants retain the authority and capacity to interrogate, override or contextualise algorithmic outputs, backed by independent auditing of model performance, bias and outcomes and enforceable appeal and redress rights.

  55. Clarify liability for AI harm involving multiple vendors, developers and government agencies.

    Clear liability frameworks are needed to determine responsibility when harm arises, particularly in systems involving multiple actors such as vendors, developers and government agencies.

  56. Ensure public institutions can test, verify and audit the AI systems they deploy without relying solely on vendor claims, with oversight shared between the Attorney-General's Department, the Commonwealth Ombudsman and the OAIC.

    Public institutions must be able to test, verify and audit the AI systems they deploy without relying solely on vendor claims or proprietary processes, with oversight shared between the Attorney-General's Department, the Commonwealth Ombudsman and the Office of the Australian Information Commissioner.

  57. Run election administration, parliamentary communication, public service announcements and government digital services on infrastructure governed by Australian law.

    Election administration, parliamentary communication, public service announcements and government digital services must operate on infrastructure governed by Australian law

  58. Establish sovereign verification capability for synthetic media: national digital signature infrastructure for public officials, AI-enabled synthetic media detection tools, and public-facing verification portals.

    sovereign verification capability becomes essential: national digital signature infrastructure allowing public officials to authenticate their own statements, AI enabled detection tools for synthetic media, and public facing verification portals would allow citizens to distinguish authentic government communication from fabricated content.

  59. For AI materially affecting employment, safety or consumer welfare, enforce guardrails through existing sector regulators with AI Safety Institute technical assessment and existing order-making and penalty powers.

    For AI systems that materially affect employment, safety or consumer welfare, compliance with the guardrails should become a legal obligation enforced by the regulator with existing jurisdiction over that sector, such as the Fair Work Commission for workplace deployments or the Australian Competition and Consumer Commission for consumer facing systems, with the Australian AI Safety Institute providing the technical assessment those regulators currently lack the capacity to perform themselves. Deployments found not to meet the guardrails should be subject to the same order making and penalty powers the relevant regulator already holds for comparable non-compliance in its sector, rather than a new and separate enforcement regime being built from scratch.

  60. Align sovereign AI obligations with critical-infrastructure and national-security frameworks such as the Security of Critical Infrastructure Act, rather than assuming AI risks are covered.

    Existing frameworks such as the Security of Critical Infrastructure Act were not designed with AI model assurance in mind. Aligning sovereign AI obligations with these frameworks, rather than assuming they already cover AI-specific risk, is an immediate task, not a completed one.

  61. Enact an AI Accountability Act defining developers, deployers and catastrophic risk, with high-risk human oversight and annual independent Institute reviews; pair it with PEReN-style sovereign technical audits, retaining regulators' legal authority.

    An Artificial Intelligence Accountability Act, establishing statutory definitions of developers, deployers and catastrophic risk, mandatory human oversight for high-risk systems, and annual independent reviews by the Australian AI Safety Institute, paired with a dedicated sovereign technical audit capability modelled on France's Pôle d'Expertise de la Régulation Numérique, where regulators retain legal authority while a dedicated forensic body performs the technical work

  62. Protect creative labour through strong IP protections, transparent licensing and enforceable attribution and remuneration rights.

    Strong IP protections, transparent licensing frameworks and enforceable rights to attribution and remuneration are essential to ensuring that AI supports rather than undermines creative work.

  63. Give regulators powers and capability to audit training datasets, enforce copyright, regulate automated publication and protect Indigenous cultural knowledge from unauthorised use.

    Finally, regulatory institutions must have the authority and capability to oversee AI’s impact on intellectual property and cultural content. This includes the ability to audit training datasets, enforce copyright compliance, regulate automated content publication and protect Indigenous cultural knowledge from unauthorised use.

  64. Build sovereign synthetic-media detection, including cross-platform monitoring, rapid disinformation response and public verification tools.

    Dedicated sovereign detection capability is required including cross-platform monitoring for coordinated manipulation, rapid response mechanisms for disinformation incidents, and public-facing verification tools, so that Australia is not dependent on offshore systems to identify harms occurring within its own information environment.

  65. Enable public officials and institutions to authenticate statements and announcements through cryptographic signatures and secure identity verification, with tamper-resistant communications.

    Public officials and institutions must be able to authenticate their own statements, video, audio and public announcements through mechanisms such as cryptographic signatures and secure identity verification, paired with tamper resistant communication channels.

  66. Embed cybersecurity in infrastructure planning, mandate independent verification of essential-service AI and build national red-teaming capability for allied AUKUS Pillar II testing.

    Ensuring resilience requires embedding cyber security into infrastructure planning, mandating independent verification of AI systems used in essential services, and building a national AI red-team capability able to participate in allied red-teaming and threat testing programs under AUKUS Pillar II.

  67. Create structured channels for telecommunications, media and platform-adjacent workers to report algorithmic anomalies or manipulation to relevant authorities.

    Workers in telecommunications, media and platform adjacent industries have direct operational visibility into how content is moderated and how platform governance obligations are met in practice rather than on paper, and structured channels for reporting algorithmic anomalies or manipulation patterns to relevant authorities would convert that frontline knowledge into a genuine early-warning function that no regulatory body currently has equivalent access to.

  68. Give the Fair Work Commission AI panel statutory authority for independent advice, complaints oversight and binding dispute resolution on workplace AI and algorithmic management.

    In workplace governance, the Fair Work Commission would need to be expanded to integrate AI oversight within Australia's industrial relations system directly, supported by a dedicated Expert Panel on AI with statutory authority to provide independent advice, oversee complaints and exercise binding dispute-resolution powers over workplace AI and algorithmic management.

  69. Extend ACMA powers to require recommendation-system disclosure, independent high-risk algorithm audits and mandatory timely enforceable reporting of coordinated inauthentic behaviour and foreign interference.

    In platform and information governance, ACMA's mandate would need to extend beyond its current information request powers to require platforms to disclose how recommendation systems operate, allow independent audit of high-risk algorithms, and meet mandatory, timely and enforceable reporting obligations covering coordinated inauthentic behaviour and foreign interference attempts.

  70. Legislate annual independent Institute reviews of developers' risk management and human oversight, public findings, worker review and input, and mandatory annual developer/deployer reporting under an accountability framework.

    The Institute should be legislated to conduct annual independent reviews of developers' risk management systems and the efficacy of human oversight mechanisms, with findings reported publicly and with provisions enabling workers to review and provide input on system performance and impacts – a function that gives the Institute both technical depth and a direct line to how AI systems actually behave in workplaces, not just in laboratory testing. Complemented by mandatory annual reporting from developers and deployers under a legislated accountability framework,

  71. Integrate AI governance into security agencies' operational frameworks.

    The security community, including defence, intelligence, cyber agencies and emergency-management organisations, must integrate AI governance into their operational frameworks.

  72. Create rapid cross-jurisdiction information sharing and a centrally coordinated cross-portfolio register of AI dependencies, vulnerabilities and emerging threats.

    They also need mechanisms to share information rapidly across jurisdictions. A cross-portfolio register of AI dependencies, vulnerabilities and emerging threats, coordinated centrally, would ensure

  73. Extend union participation in the Sovereignty Compact to co-design deployment conditions and provide an early-warning function based on frontline system behaviour.

    Union participation in the Compact should extend beyond workforce representation to include co-design of deployment conditions and a genuine early warning function, since workers in telecommunications, media and critical infrastructure have frontline visibility into how systems behave under real conditions that regulators and technical audits alone cannot fully replicate.

Read the original

Open in the interactive guide